This fixes issue #495.

The sub_t.subnum's that were being updated as message areas were being saved to
msgs.ini could not be used as an index into the scfg_t.sub array at this point
(the subnum would be the new index position when the msgs.ini was re-read/loaded)
.

This was not an issue in v3.19 because we just saved the subnum (to msgs.cnf) and
in v3.20, we save the sub's internal code (to msgs.ini) and were using the newly
updated sub_t.subnum to find the corresponding sub_t for that internal code. Since
the subnum is not used now during the save process, no need to update it here
(this reverses part of the commit 11e529d4 from 5 years ago).

This fixes issue #502 - thanks to the irc.synchro.netizens that reported it!

IF user hits 'Q' (or whatever the "Quit" key is), set the default protocol
field in the user record to " " (instead of an empty string).
If user hits abort (Ctrl-C), don't make any change to the default protocol.

A blank download protocol field in a user.dat, when parsed, sets the 'prot' field
of user_t to 0. When writing the record back to the user.dat, this would prematurely
truncate all other fields off the user record (since strings in C are NUL terminated
and we're using sprintf() to format the record and %c specifier for that field).

The fix is to write a ' ' character instead of '\0' if the user_t.prot is '\0'.
As part of this fix, I'm writing a '?' if a user_t.sex is '\0' (not sure if this
is actually possible, but just as insurance). Those are the only 2 single-character
user properties/fields today.

Bug reported/debugged by Al of The Rusty Mailbox (1:153/757.2) - thank you!

This function has been causing somewhat rare crashes (e.g. one per week on a very active system with MQTT enabled) on both Windows and Linux (see issue #495). The root-cause is still unclear (possible heap corruption?). This function needed more robustification anyway (see previous TODO comment), so I'm refactoring here to no longer use strListAppendFormat, which uses vasprintf (the function where the rare crash was occurring) and instead just use snprintf and strListPush. The total client_list that's published (as a single string/message) is now dynamically allocated (this was the point of the previous TODO comment). This may not actually fix the issue if there's a heap corruption occurring somewhere else in this function or the call-chain that's reaching here (usually from the web server).

Another change to mqtt_client_on(): don't incremented the server's 'served' counter until client disconnects (this is a past-tense statistic).

Also:
- mqtt_pub_strval() can now accept a null 'str' argument which is effectively the same as mqtt_pub_noval(), so the pub_noval() function is now just a thin wrapper around pub_strval().

- mqtt_startup() now clears the client_list topic and sets the client_count topic to 0.

Reduce occurrences of subsequent errors (unsure of root-cause):
!ERROR 2 (No such file or directory) in main.cpp line 2778 (event_thread) removing "/sbbs/data/pack0341.now"

This does seem to be the only location that deleted pack*.now files, so weird that they'd disappear sometime after calling glob(), but I do understand that its theoretically possible.

No immediate need/use.

to remove all sections in an .ini file or all sections with a specified
prefix.

Allows for more/better wildcard matching in viewable and testable
file types and download-events (e.g. "tar.gz" or "tar.*").

This solves the "double-dot" problem in some file types/extensions.
An implicit "*." is prepended before the configured file type/extension.

If libarchive couldn't extract a QWK or REP packet, we'd fallback to searching
for a match among the configured extractable file types and if no extension/type
match was found, default to the first configured extractable file type
(even if there wasn't one) which would result in a NULL pointer dereference
and most likely a crash. Instead, if no matching configured extractable
file type is found, just log a warning message and don't continue with the
extraction attempt. With SBBS v3.19+, it's totally valid/legit to have no
extractable file types configured in SCFG and things "just work" (using
libarchive).

For Viewable, Testable, Compressible file types and download events,
the file extension is no longer limited to 3 characters. 15 characters
seemed like "enough".

Similarly, a user's temp/QWK file type preference is also extended from
3 to 15 characters.

This fixes issue #486, but file types such as *.tar.gz (2 dots in the extension)
are still an issue and will be addressed next.

... but the old user.dat fields would still just be 16 chars in size.

Also a strcpy -> SAFECOPY conversion for user tmpext.

I read that 3.1.1 is much more widely supported, but perhaps that was old
info.

When MQTT protocol version is >= 5, just log to the */log topic
(the level is a property).

This means you can't subscribe only to specific log levels when using MQTT v5
so I may re-think this, but I don't want to keep publishing every single log
message twice.

Value is "http" or "https" depending on if TLS is in use.

This can be useful when debugging issues (e.g. crashes) and an MQTT consuming/logging client wasn't connected at the time.

Yes, someone could have a local hostname of "node". <sigh>

This solves the problem of subscribing to "sbbs/+/+/+" and getting node status in addition to server status (very different formatted messages).

Also include a change for the singular-again "node" topic

If a configuration had *no* command shells, this would have caused a crash here in SCFG.

Also, when selecting the new user command shell, default to the current selection.

Should fix the segfault reported by NeoArata_ in #synchronet

Also include singular/plural error/s in server status messages.

<server>/state is now published directly to the <sever> topic level (the "state" topic goes away).

Moved <bbs-id>/node# topics to <bbs-id>/nodes/# (allows better wild-card subscription per MQTT standards).
Publishing total node count to <bbs-id>/nodes rather than <bbs-id>/node_count

Introduced "publish verbosity" control (defaults to "High") - if you want nice human readable (and sometimes redundant) topic messages, leave this set to "High". To reduce traffic, set to "Low". It's expected that purpose-built Synchronet/MQTT clients should work equally-well with either setting, but when using generic MQTT clients/browser, "high" verbosity is nice. The human-readable node status is only published when "high" verbosity is enabled.

Publishing the BBS name to the <bbs-id> topic, the instance hostname (as configured in sbbs.ini) to the <host> topic. Makes the hierarchy much more clear when using an MQTT browser like MQTT Explorer.

This is more consistent with how these events are logged in a BBS-common log file in data/*.log.

This change also restores the server abbreviation to the error log entries that used to be there until recently.

Makes for a nice node listing with pure-text MQTT clients/tools

Each Synchronet server is now its own MQTT client. This means there's no
longer any MQTT logic in the Synchronet "hosts" (e.g. sbbscon.c, ctrl/*.cpp)
and none needed for SBBS NT services (they'll "just work" with MQTT).

This also means that just about everything (except for nodes, spam and hack)
is now published per-server (in the sbbs/BBS-ID/hostname/server/ topic branch)
and if you want aggregated totals or client lists, you'll have to do that in
your own MQTT client or dashboard.

I also removed the publishing of thread_count and socket_count topics as
they weren't universally supported across all servers and are of questionable
value. They can be added back later if determined to be useful.

A set_state() may be called with SERVER_STOPPED while recycling.

any server "running" is not enough because the server could still be
initializing, re-initializing, or even shutting down.

This solves the problem of any signal (even HUP) ungracefully terminating
sbbs (on *nix) when MQTT was enabled. As reported by Nightfox on IRC.