refactor(posixage/flock): tidy tryLock structure and test name

[Benehiko]

Clean up follow-ups to the ctx-driven lock-wait refactor (#523):

• Collapse the two sequential ctx.Err() checks into a single early-return gate before stale recovery. Equivalent behavior with a more linear read; the ctx is also checked again inside retryLock via backoff.Retry so the explicit re-check after recovery was redundant.

• Clear fl immediately after calling recoverStaleLock so the outer close-on-error defer does not double-close the file handle (which recoverStaleLock already closes via its own defer).

• Rename the test "caller context can wait past former default timeout" to "acquires lock when holder releases before ctx deadline" — the former-default-timeout reference no longer applies.

[docker-agent]

Assessment: 🟢 APPROVE

The refactoring is clean and correct:

• Early ctx check: Moving ctx.Err() before recoverStaleLock correctly skips unnecessary file I/O when the context is already cancelled.
• Double-close fix: Setting fl = nil unconditionally after recoverStaleLock(root, fl) correctly prevents the deferred close-on-error from double-closing a handle that recoverStaleLock already closes via its own defer. This is a genuine improvement over the previous code.
• Removed second ctx check: Safe — retryLock respects context cancellation internally via backoff.Retry(ctx, …). The only subtle behaviour change is that a cancellation occurring in the narrow window between the early-exit check and retryLock now returns an error wrapping ErrLockUnsuccessful (via retryLock), whereas the old code returned a bare ctx.Err() for that specific path. This is arguably more consistent, and all existing tests already assert errors.Is(err, ErrLockUnsuccessful).
• Test rename: Cosmetic improvement — the new name better describes the scenario being tested.

No resource leaks, nil dereferences, or logic errors found.