DailyStatsActor: move SQLite EnsureTable off PreStart hot path

[Aaronontheweb]

Summary

DailyStatsActor.PreStart() performs synchronous SQLite I/O — connection
open + two CREATE TABLE IF NOT EXISTS statements + commit (see
src/Netclaw.Daemon/Gateway/DailyStatsActor.cs:105-110, 391-435). The
`QuerySkillUsageStats` handler opens a second SQLite connection. On
Windows CI cold-start, AV / Defender scanning the freshly-created `.db`
file plus first-time fsync on COMMIT regularly pushes the round-trip
beyond the test's 3-second `Ask` timeout.

Symptom

Netclaw.Daemon.Tests.Gateway.DailyStatsActorTests.QuerySkillUsageStats_returns_groupable_rows_for_each_method
flakes on Windows CI with `Akka.Actor.AskTimeoutException : Timeout
after 3.00 seconds` at `DailyStatsActorTests.cs:44`. Linux passes
consistently. Same SHA can pass and fail on consecutive reruns —
genuinely flaky, not deterministic.

Observed once on PR #916 rerun
(`https://github.com/netclaw-dev/netclaw/actions/runs/25523848615\`).

Root cause analysis

Both akka-net-specialist and dotnet-concurrency-specialist agents
agree: this is a latency issue, not a race. The actor's mailbox FIFO
guarantees the four `Tell`s before the `Ask` are correctly
ordered. The latency comes from disk I/O on the actor's startup path,
not from any synchronization race.

Latency contributors on Windows CI cold start:

• `PreStart`: SQLite connection open + transaction + 2 CREATE TABLEs
  • COMMIT (fsync). First-time file creation triggers AV scan.
• Query handler: a second SQLite connection open + SELECT, with
  Microsoft.Data.Sqlite pool keyed by connection string (which is
  unique per test, so pooling does not help).
• xUnit parallelism on a contended Windows runner amplifies both.

Proposed fixes (priority order)

1. Move table creation off the actor's startup path. Run
   `EnsureTable()` once at `ActorSystem` setup or as part of the
   existing `SchemaMigrationHostedService`. The actor should not
   block its mailbox on schema DDL. This eliminates the latency at its
   source for both tests and production startup.

2. Pre-warm in tests. Open and close a SQLite connection plus run
   the CREATE TABLE statements before `ActorOf` in the test fixture.
   Test-only mitigation; does not help production cold start.

3. Cheap drive-by: bump the test's `Ask` timeout from 3s to 10s.
   Hides the slow path but does not fix it. Worth shipping immediately
   to unblock CI; do (1) as the proper fix.

Acceptance criteria

• DailyStatsActor's `PreStart` no longer performs SQLite DDL.
• The test passes with a tight (≤5s) `Ask` timeout on Windows CI
  reruns.
• Schema initialization runs deterministically before the actor
  system starts processing messages.

Reference

• Failing run: https://github.com/netclaw-dev/netclaw/actions/runs/25523848615/job/74915830222
• Source: `src/Netclaw.Daemon/Gateway/DailyStatsActor.cs:105-110, 391-435`
• Test: `src/Netclaw.Daemon.Tests/Gateway/DailyStatsActorTests.cs:30-64`