CloseCallbackBudget can skip internal PreShutdown hooks (resync stop, LimitFilter cleanup)

[benitogf]

Problem

Server.Close runs registered close hooks under an optional CloseCallbackBudget: once a positive budget is exhausted, runPhase skips the remaining hooks in that phase. Internal, lifecycle-critical hooks are registered through the same mechanism as user callbacks, so a tight budget can skip them:

• the resync worker's stopResync (PreShutdown), added in fix(stream): make the broadcast the only writer of the pool cache #150 — a skipped stop leaves the worker goroutine alive into Storage.Close, i.e. a goroutine leak and a possible read-after-close on embedded storages.
• LimitFilter.StopCleanup (PreShutdown, filters.go) — same exposure, pre-existing.

With the default budget of 0 nothing is skipped, so this is latent, not a live bug. But mixing internal teardown with budgeted user callbacks means an operator who sets a budget to bound user callbacks can silently break internal shutdown.

Suggested direction

Exempt internally-registered lifecycle hooks from CloseCallbackBudget — e.g. a separate non-budgeted registration path (or a flag on the hook) so the budget only ever caps user-supplied callbacks. Internal teardown should always run.

Surfaced during the self-review of #150; out of scope there.

🤖 Generated with Claude Code

[benitogf]

Fixed by #153 (fix(close): exempt internal teardown hooks from CloseCallbackBudget), merged to main. Internal teardown hooks (stopResync, LimitFilter.StopCleanup) are now exempt from CloseCallbackBudget, so a tight budget can no longer skip them. Closing.