[On1x]

The file appender used to schedule the next rotation with fc::async/fc::schedule on the
fiber queue of the thread that constructed it. vizd builds its loggers in main() and then
lives in boost::asio::io_context::run(), so that queue is never serviced again: the initial
rotation ran, the next task sat forever, and the node kept writing into a single file (measured
2026-10-20: 37 MB in 9 h on testnet, no new hourly file in 4 days on mainnet). While there,
_current_file_start_time was not updated in the "same hour" branch either.

Bumps thirdparty/fc to VIZ-Blockchain/fc@3968605 (merged into fc master): rotation is now
serviced by a dedicated std::thread (same approach as the snapshot plugin) — start()
performs the initial rotation synchronously, the timer sleeps in <=1 s slices until the next
boundary, shutdown joins the thread, and timer exceptions back off 1 s instead of terminating
or spinning. _current_file_start_time is written under slock.

No consensus impact: this is logging infrastructure only.

Verification (container on the project's boost-base-1.91 image, harness with main-thread
sleeps, i.e. the same threading picture as io_context::run()):

• new code, 1 s interval, 4 s: 6 files, 40/40 lines, no losses or duplicates;
• old code, same stand: 0 files without an fc pump, 2 files with one startup pump (the prod picture);
• 3600 s interval: no spurious rotation.

Docs updated on all three locales: rotation is now serviced by the appender's own thread, and
binaries built before 2026-10-25 still need external logrotate.