packer: don't panic on Put after Wait

[poolpOrg]

Summary

• Fix PlakarKorp/plakar#2106: backup interrupted with Ctrl+C (or any error path that calls `PackerManager.Wait` while workers are still calling Put) can panic with `send on closed channel` in `seqPackerManager.Put`.
• Root cause: Wait closed every `packerChan[i]`; Put sent to them unconditionally. The two race.
• Fix: replace the close-on-Wait with a dedicated `closing chan struct{}`. Put `select`s between sending and observing closing, returning `ErrShutdown` on shutdown. Run workers also observe closing and flush their tail packfile before returning. Wait is idempotent via `sync.Once`. `packerChan` itself is never closed, so no possible "send on closed channel".
• Same shape of fix applied to `platarPackerManager` (.ptar) for consistency.

What about the lifecycle bug in snapshot.Backup?

There's an upstream issue too: `importSource` returns early on `importerErr` without waiting for `<-errch`, so workers can still be calling `Put` when `Backup` calls `PackerManager.Wait`. That's a separate fix worth doing on the plakar / snapshot side, but the packer should also not panic regardless of who calls Put with what timing — which is what this PR addresses.

Test plan

• New `TestSeqPackerPutDuringWait` races 32 concurrent producers (200 puts each) against a Wait. Hits "send on closed channel" reliably on `main`; passes on this branch under `-race -count=50`.
• New `TestSeqPackerPutAfterWait` pins the post-shutdown contract: Put returns an error rather than panicking.
• Full `go test ./...` clean on kloset.
• Full `go test ./...` clean on plakar (built against this branch via a local replace).
• End-to-end: 5 backup runs of ~100MB each, all SIGINT'd mid-flight. 0 panics. (On `main` this is the reproducer pattern from the bug report.)

(There are two pre-existing `-race` failures in the packer test suite — `TestPackWriter_ConcurrentInternalAccess` and `TestPlatarPackerManager_RunAndWait` — both reproducible on `main` without this change. Out of scope here.)

🤖 Generated with Claude Code

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review to trigger a review and subscribe this PR to future pushes, or @claude review once for a one-time review.

_{Tip: disable this comment in your organization's Code Review settings.}

[poolpOrg]

Fixes an actual bug, analysis is sound, PR seems ok after an initial review, but I need to discuss this again with @mathieu-plak as I'm not fully convinced about the draining bits.

[mathieu-plak]

#376 big conflict energy :).