fix: repair missing action rows on status update

[Rohithmatham12]

Summary

• return a repository-level ErrActionNotFound when UpdateActionPhase matches no row and the action truly does not exist
• map that sentinel to a NotFound status in InternalRunService.UpdateActionStatus
• make the actions watcher re-record the TaskAction and retry UpdateActionStatus when the run service reports NotFound
• preserve stale/backward phase no-op behavior by checking row existence before returning NotFound

Fixes #7257.

Testing

• GOCACHE=/private/tmp/flyte-go-cache GOMODCACHE=/private/tmp/flyte-go-mod-cache go test ./actions/k8s -run 'TestNotifyRunService_RecordAndRetryWhenStatusUpdateReturnsNotFound$'\n- GOCACHE=/private/tmp/flyte-go-cache GOMODCACHE=/private/tmp/flyte-go-mod-cache go test ./runs/service -run 'TestUpdateActionStatus_MissingActionReturnsNotFoundStatus$'\n- GOCACHE=/private/tmp/flyte-go-cache GOMODCACHE=/private/tmp/flyte-go-mod-cache go test ./runs/repository/impl -run 'TestUpdateActionPhase_(MissingActionReturnsNotFound|BlocksBackwardFromSucceeded)$'

[AdilFayyaz]

Any database error is reported as "not found". GetAction returns an error for timeouts and connection problems too. So those become CodeNotFound and trigger an unneeded re-record.

[AdilFayyaz]

RecordAction failures inside the response are treated as success. We should catch the resp here as well and then check resp.Msg.GetStatus().GetCode() to log failures.

[Rohithmatham12]

Rebased on current main and addressed the review feedback.

What changed:

• Preserved the upstream in-band UpdateActionStatus rejection handling while keeping the missing-action retry path.
• RecordAction now treats non-zero resp.Status.Code as a failed record, logs it, and does not add the action to the recorded filter.
• UpdateActionPhase now only converts the fallback lookup to ErrActionNotFound when the action is truly missing; lookup/DB errors are returned as internal verification failures instead of being misreported as not found.
• Added regression coverage for retrying RecordAction when the run service rejects the record in-band.

Validation:

• git diff --check
• GOCACHE=/private/tmp/flyte-go-cache GOMODCACHE=/private/tmp/flyte-go-mod-cache go test ./actions/k8s -run 'TestNotifyRunService_(RejectedStatusUpdateSkipsTerminalLabel|RecordAndRetryWhenStatusUpdateReturnsNotFound|RetriesRecordActionWhenRunServiceRejectsRecord)$'
• GOCACHE=/private/tmp/flyte-go-cache GOMODCACHE=/private/tmp/flyte-go-mod-cache go test ./runs/service -run 'TestUpdateActionStatus_MissingActionReturnsNotFoundStatus$'

I also attempted the focused repository DB tests, but the local embedded Postgres test port was already occupied on this machine:

failed to start embedded postgres on port 15432: process already listening on port 15432

That package should get a clean DB process in CI.