Fix memory-safety and correctness bugs surfaced by Coverity audit (part 8) - #22293

Summary

Single follow-up commit not yet covered by parts 1-7 (#22266, #22267, #22268, #22270, #22279, #22280, #22281).

@stelfrag — could you pick this one up and review when you get a chance? Thanks for splitting the original PR into reviewable parts.

CID 442107 (SLEEP) — daemon: free removed stale nodes outside rrd lock

remove_ephemeral_host() in src/daemon/commands.c held the global RRD write lock while calling rrdhost_free___while_having_rrd_wrlock(), which performs blocking teardown work (logging, stream-shutdown waits, data collection cleanup, health cleanup). Every other RRD writer was stalled meanwhile.

The fix splits host removal into:

• rrdhost_unlink___while_having_rrd_wrlock() — fast index removal, runs under the lock
• rrdhost_free_unlinked() — slow teardown (cleanup_data_collection_and_health, free), runs after the lock is released

A new wrapper rrdhost_free___without_having_rrd_wrlock() encapsulates the lock-then-unlink-then-unlock-then-free pattern for callers like remove_ephemeral_host().

Files changed:

• src/daemon/commands.c
• src/database/rrdhost.c
• src/database/rrdhost.h

3 files changed, +23 -15.

Test plan

• CI build passes
• Coverity scan no longer reports CID 442107

Summary by cubic

Fixes Coverity CID 442107 by moving slow host teardown outside the global RRD write lock to prevent stalls. Host unlinking now happens under the lock, with teardown after unlock for better responsiveness.

• Bug Fixes
  • Unlink under lock with rrdhost_unlink___while_having_rrd_wrlock(), then teardown via rrdhost_free_unlinked() after releasing the lock.
  • Added rrdhost_free___without_having_rrd_wrlock() and updated remove_ephemeral_host() to use it.
  • Keeps cleanup order intact while minimizing lock hold time.

^{Written for commit 509be03. Summary will update on new commits. Review in cubic}