fix(event): stop the quoting sink from swallowing every audit channel

[esengine]

Found by running reasonix run --trajectory for real while verifying #8404, not
by a test — every test in the tree wires recorders directly and stayed green.

CostQuoteSink forwards Emit but not the optional sink capabilities, so every
recorder below it in a chain receives nothing. In the CLI run chain that is the
stats recorder and the trajectory recorder, both of which sit under it:

inboxEventSink        capable=true
goalUsageTee          capable=true
event.coalescer       capable=true
event.syncSink        capable=true
event.CostQuoteSink   capable=false   <-- chain ends here
stats.Recorder        capable=true    (never reached)
trajectory.Recorder   capable=true    (never reached)

A real research run wrote 2130 event records and zero audit records: no
outcome progress, no readiness audits, no completion reports, no delegation
admissions, no protocol recovery, no workspace mutations.

That matters beyond diagnostics. Those channels carry the shadow samples the EBM
and reasoning-governor experiments were built to compare (ebm_eligible,
governor_engaged, debt_age, blind_mutations), so CLI trajectories have been
empty of them, and cmd/e2ebench has been silently taking its
"older recordings" fallback — the one its own comment notes cannot price scorer
claims — instead of reading recorded samples.

The fix is to embed AuditForwarder, which exists for exactly this case and
whose comment already records the lesson: hand-written forwarding lost
capabilities at multiple wrappers even while their owning tests stayed green.
The quoting sink simply never adopted it.

The regression test drives the package-level dispatchers rather than calling the
methods directly. That matters: the dispatchers are the path the host actually
uses and they no-op silently on a wrapper that lost the capability, so the test
fails with a named channel per capability. Calling the methods directly would
only produce a build error, and only if they vanished outright.

Verified end to end: with the fix, the same reasonix run --trajectory now
records outcome_progress.

Cache-impact: none - sink capability plumbing only; no tool schema, prompt, or provider request surface is touched, so the system-prompt prefix stays byte-identical.
Cache-guard: go test ./internal/boot/ (provider_request golden), unchanged and green.
Documentation-impact: none - no user-facing behaviour or documented contract changes; this restores recording that the docs already assume happens.