fix(flowcontrol): map pre-admission TTL/cancel to eviction statuses i…

[LukeAVanDrie]

What type of PR is this?

/kind bug

What this PR does / why we need it:

Two outcome-mapping fixes in flow control admission.

1. A TTL expiry or client disconnect that fires before an item reaches a queue (buffered in the enqueue channel, blocked in submission, or racing managedQueue.Add) finalizes as QueueOutcomeRejectedOther, which translateFlowControlOutcome mapped to 500. Once the item is queued, the same event maps to 503 with a dropped-reason header. These pre-admission windows are widest under overload, so a saturation stall with TTL expiries showed up as a spike of 500s that was really backpressure.

   The translator now recognizes ErrTTLExpired and ErrContextCancelled in the RejectedOther/EvictedOther arm (after the shutdown check, which keeps precedence) and delegates to the matching eviction mapping, so the two paths cannot drift apart. The same branches cover EvictedOther for symmetry; nothing currently emits it with these sentinels.

2. tryDistribution finalized a ManagedQueue lookup failure for a leased flow as QueueOutcomeRejectedCapacity, reporting an internal invariant violation as 429 with the Saturated header. It now finalizes as RejectedOther: a 500 with no saturation header. This also keeps the RejectedCapacity label in flow_control_requests_total to genuine capacity rejections. The registry error is flattened with %v because wrapping it would let ErrPriorityBandNotFound flow back through the connection closure and trip the priority-0 fallback in withConnectionWithFallback.

Coordination with #2047: it changes the same function (queued TTL eviction becomes 429 when the pool has endpoints). Whichever lands second threads the ttlPoolEmpty parameter through the two delegating calls, and the pool-emptiness probe in Admit should also cover RejectedOther + ErrTTLExpired so pre-admission TTL gets the same 429/503 split.

Tests: new TestTranslateFlowControlOutcome rows for pre-admission TTL and cancel on both outcomes, plus a shutdown-precedence row. The controller lookup-failure test now also asserts that registry sentinels do not leak into the finalized error and that no fallback retry happens. Both packages pass with -race.

Which issue(s) this PR fixes:

Fixes #2097

Part of #1187.

Release note:

NONE

[LukeAVanDrie]

I would like to hold this until #2047 is merged upon which I will make the updates mentioned in the PR description.

[ahg-g]

I would like to hold this until #2047 is merged upon which I will make the updates mentioned in the PR description.

merged now