relay: a reaction on a project root (issue/PR) panics the ingest worker

[LiterallyBlah]

Bug: a reaction on a project root (issue/PR) panics the relay's ingest worker

Component: buzz-relay — crates/buzz-relay/src/handlers/ingest.rs, reaction write path
Severity: crash + client livelock (stored-then-panic makes retries permanent)

Symptom

Publishing a NIP-25-style reaction whose target is a project root or project comment (kind 1621 issue, 1618 PR, or a kind-1 comment on one) panics a tokio worker:

thread 'tokio-rt-worker' panicked at crates/buzz-relay/src/handlers/ingest.rs:2824:51:
reaction path has channel

Root cause

The reaction path's conformance-trace emission asserts a channel is always present:

// This branch is the reaction path; channel_id is always Some here, so
// WriteInsertGlobal does not apply.
channel: channel_label(channel_id.expect("reaction path has channel")),

That assumption predates reactions on project events. Project events carry no h tag, so channel_id is None — the expect panics. Two aggravators:

1. Stored-then-panic: the reaction row is inserted before the trace action is built, so the client sees a failed request for an event that was actually persisted, and retries.
2. The duplicate branch carries the identical expect, so every retry panics too. A client with a durable, ordered publish queue (e.g. buzz-acp's gated-publish FIFO acknowledging a project dispatch) is head-of-line blocked forever, burning rate-limit quota on a publish that can never be acknowledged.

Any human tapping a reaction on an issue comment in Desktop triggers the same panic.

Fix that worked for us

Mirror the message write's three-way split at the same seam (it already handles channel-less writes a few hundred lines below): (Some, true) → WriteInsert, (Some, false) → WriteDuplicate, (None, _) → WriteInsertGlobal. The conformance vocabulary already models channel-less writes; only the reaction path was missing it.

Reproduction

1. Closed or open relay; publish a kind-1621 issue.
2. React (kind 7) to the issue root or a comment on it, with no h tag (as Desktop/agents correctly send for project events).
3. Observe the panic in relay logs; observe the reaction IS stored; retry and observe the duplicate-branch panic.