fix: surface swallowed failures on data-library indexing and publish botId lookup (OBS follow-up) @W-24016300

[nico-pappagianis]

Summary

Follow-up to #351 (logging standardization) that implements the three behavioral observability findings deferred out of that PR's review (OBS-1 / OBS-2 / OBS-5). #351 deliberately kept to a pure logging-format change; this PR makes the small behavior changes those findings actually require.

Draft / stacked on #351. Base is chore/standardize-logging, not main, because these changes build on the logCtx helper introduced in #351. Rebase onto main once #351 merges.

What changed

OBS-2 — getPublishedBotId no longer collapses every error into "not published"

singleRecordQuery throws SingleRecordQuery_NoRecords when zero rows match — that is the genuine "agent not yet published" case (→ debug, return undefined, POST a first version). Any other error (auth, network, multiple matches) means the lookup itself failed and we cannot tell whether the agent exists. Previously all of these collapsed to "not published", risking a duplicate first version POSTed over an existing agent. They now warn, emit agent_publish_botid_lookup_failed telemetry, and rethrow.

OBS-5 — KNOWLEDGE indexing trigger no longer swallows real failures

The fire-and-forget indexing fetch had an empty catch {}. An intentional 10s AbortError stays debug (the server processes indexing asynchronously on receipt, so a timeout is expected and harmless). A genuine failure — the request never reached the server — now warns, so a data library that silently never indexes is diagnosable.

OBS-1 — sole-signal swallow paths promoted above the default WARN capture

Logger.DEFAULT_LEVEL is WARN, so debug lines don't persist in a default production capture. OBS-1 asked that any path which is the only signal of a problem and swallows-and-continues log above that threshold. After OBS-2 and OBS-5, those were the only two such paths, and both now warn on genuine failure. The log-and-rethrow catches (agentDataLibrary, apiCatalog) correctly stay at debug — the throw plus its _failed telemetry is the durable signal, so promoting them would be log-and-throw double-reporting.

Tests

• should return undefined when agent does not exist now simulates not-found with the realistic SingleRecordQuery_NoRecords SfError instead of a generic Error.
• New: should rethrow when the lookup itself fails (not a genuine not-found).
• Publish-error tests that relied on the old swallow behavior to reach the POST now stub the not-published lookup explicitly.

yarn tsc -b clean; yarn mocha → 420 passing, 1 pending.

Open question for review

OBS-2 changes a real lookup failure from swallow-and-continue to rethrow. That is the safer default (don't risk a duplicate publish when org state is unknown), but it does turn a previously-silent transient query blip into a hard publish failure. Flagging in case we'd rather warn + telemetry and still proceed. Draft pending that call.

@W-24016300

[setu4993]

Nice fixes. Let's make sure CI is all green before we merge.