fix(poster): eliminate freeze/hang on large file batches

[javi11]

Summary

Fixes the long-standing freeze reported in #168 where uploads hang overnight under large file counts and only recover after a Docker restart. The reporter notes v0.0.29-rc5 was the last stable release; every release since has the bug.

A code-level audit of internal/poster/poster.go revealed several concurrency and resource-management defects that match the symptoms (silent freeze under volume + time, no panic surfaced). All are fixed in this PR with minimal-diff edits — no architectural rework.

Root causes & fixes

#	Bug	Fix
1	postQueue closed unconditionally right after the initial enqueue loop. Later, checkLoop's retry send select { case postQueue <- failedPost: ; default: } could not catch a closed-channel send and panicked. With no recover(), the goroutine died and wg.Wait() blocked forever.	Gate the close on postsInFlight.Wait() (the WaitGroup was already wired but never waited). Rebalance the retry path so the original post is Done'd when its retry is queued.
2	post.file was not closed on postLoop pool-error or verify-exhausted-no-deferred paths. Every failed file leaked an fd; long-running daemons hit the ulimit and stalled on the next os.Open.	Close post.file on every terminal path.
3	The deferred-check errChan send was the only one without a ctx.Done() companion — buffer 1 with multiple writers could deadlock the goroutine.	ctx-guarded select and bump buffer to 4.
4	Per-post read-ahead goroutine observed only the parent ctx. On the deferred-check non-fatal path the parent stays alive, so stragglers could linger.	Per-post context.WithCancel cancelled after pool.Wait().
5	failedPosts was an int incremented from checkLoop and read from the main goroutine without synchronization.	atomic.Int64 (and Post.failed accordingly).
6	Misleading default: branch in the retry select claimed to handle a closed channel — sends on closed channels always panic, default cannot intercept.	Removed; with the gating fix the branch is unreachable.

Test plan

• go build ./...
• go vet ./internal/poster/...
• go test -race -count=1 ./internal/poster/... ./internal/processor/...
• Manual soak: large directory upload overnight; verify completion and flat fd count.

Notes for the reviewer

• The nntppool v2 → v4 upgrade (feat: update nntppool dependency to v4.3.0 and refactor related code #130) and other post-rc5 changes are explicitly out of scope. If hangs persist after this lands, the pool layer is the next place to audit.
• The retry path now performs postsInFlight.Done() for the original post on successful retry-send. Without this, postsInFlight.Wait() would never reach zero and postQueue would never close.