Make a failure and the build that produced it visible in the logs

[ppXD]

Summary

A pod answered /api/auth/sign-in with 42703: column t.owner_user_id does not exist — a column #1379 renamed, which only pre-#1379 code asks for — while the latest build had been pushed. Diagnosing it took three rounds because of three separate gaps, each fixed here.

Nothing recorded which build a pod was running. Every event now carries Build (from AssemblyInformationalVersion, which the SDK stamps with the commit sha), and the first line of each boot names it, the environment, and the log destination.

The searchable copy went nowhere. Serilog:Seq:ServerUrl shipped as http://localhost:5341; every deployment inherits it, and in a container that resolves to itself. The batched sink retries against nothing and an empty Seq reads as silence. The base file now ships blank — console-only, and it says so at startup — with the zero-config local Seq moved to appsettings.Development.json. Also adds ReadFrom.Configuration + Serilog:MinimumLevel, so levels are tunable without a rebuild.

Failures thrown outside the MediatR pipeline reached no sink. GlobalExceptionFilter logged nothing, reasoning that RequestFailureObserver already had — but that only sees exceptions escaping a handler. Anything from a controller, model binding or an auth filter was a masked 500 with no line on any sink. The observer now marks what it records; the filter records what nothing else did, at the same severity.

DbUp no longer "succeeds" having found nothing. Scripts travel as content files beside the assembly; if that copy is ever missing, PerformUpgrade returns success having applied nothing and the app serves an unmigrated database — which presents exactly as this incident did.

Note on embedding the migrations

Adding <EmbeddedResource> alongside the existing content copy — the obvious hardening — would re-run all 124 migrations on any live database. DbUp journals by name, and the two providers name the same file differently:

0001_initial.sql                                       ← file system, what every DB has journalled
CodeSpace.Core.Persistence.DbUpFiles.0001_initial.sql  ← embedded, unapplied under this name

MigrationDiscoveryTests now fails if anyone adds a second script source, and the dead embedded provider (which matched nothing, and is what made the change look free) is removed.

Test plan

• FailuresAreAlwaysLoggedTests (6) — mutation-verified: filter stops logging → 3 red; observer stops marking → 1 red
• MigrationDiscoveryTests (3) — reded on the real duplicate-name hazard before it was reverted
• SeqSinkSettingsTests updated to pin both halves: base ships blank, Development ships localhost
• Unit 6474/6474 · Integration 2745/2745 · E2E 180/180