Raise the TestKit expect default for starved CI schedulers

[Aaronontheweb]

Root cause

akka.test.single-expect-default is the stock 3-second wait in every
Akka.Hosting.TestKit test. DiscordSessionBindingContractTests.Proactive_thread_can_apply_session_title_updates
flaked on Windows CI because its parameterless ExpectMsgAsync<ProactiveThreadAck>
inherits that 3-second default. The ack sits behind actor spawn,
Akka.Persistence recovery, and two stream materializations. Production
budgets 30 seconds for this same ack-after-work handshake
(ProactiveSendFormatting.ProactiveThreadAckTimeout). A starved
4-vCPU CI pool can exceed 3 seconds for a path production itself
treats as needing 30.

Reproduction

Test	Wait	Default before	Path behind the wait
DiscordSessionBindingContractTests.Proactive_thread_can_apply_session_title_updates	ExpectMsgAsync<ProactiveThreadAck>()	3s	actor spawn + Akka.Persistence recovery + 2 stream materializations
SlackThreadBackfillIntegrationTests (~L728)	ExpectMsgAsync<ProactiveThreadAck>()	3s	same class of handshake
SlackProactiveThreadTests, DiscordProactiveThreadTests, MattermostProactiveThreadTests	ExpectMsgAsync<ProactiveThreadAck>() / MattermostProactiveThreadAck	3s	same ack path
ErrorCorrelationTests	hardcoded TimeSpan.FromSeconds(3) on several waits	3s	same starved-scheduler risk on LlmSessionActor turns

Why this is an environment knob, not timeout-jiggling

Production already decided the correct budget for this handshake is 30
seconds. The test default of 3 seconds was never chosen to match the
handshake's real cost — it is the library's out-of-the-box value. This
change aligns the test default with what production already proves is
correct for the same code path, on a CI runner that is materially
slower than a dev machine. It does not touch production code, does not
loosen an assertion, and does not add a retry or a sleep. Green tests
pay zero added time because ExpectMsgAsync resolves as soon as the
message arrives — the higher ceiling only changes how long a genuinely
broken test takes to report failure.

Changes (test config only)

• SessionBindingContractTests (abstract base for the Discord/Slack/Mattermost
  session-binding contract suites) gains a Config override raising
  akka.test.single-expect-default to 15s. None of the three subclasses
  override Config, so this covers all three without touching those files.
• SlackProactiveThreadTests, DiscordProactiveThreadTests,
  MattermostProactiveThreadTests: same raise added to each
  *ProactiveThreadActorTests' existing Config override, alongside the
  existing akka.test.default-timeout = 5s.
• SlackThreadBackfillIntegrationTests: same raise added via
  builder.AddHocon(..., HoconAddMode.Prepend) in ConfigureAkka (this
  class has no Config property override to extend).
• ErrorCorrelationTests: LlmSessionTestBase.ConfigureAkka is sealed,
  so this class raises the default through the Config property instead.
  Hardcoded TimeSpan.FromSeconds(3) arguments on ExpectMsgAsync<TurnCompleted>
  calls are removed so they inherit the new default. The Ask<T> calls in
  the same file keep an explicit timeout, bumped to 15s rather than deleted
  — Ask falls back to akka.actor.ask-timeout (infinite by default), not
  single-expect-default, so deleting that argument would trade a flaky
  wait for a silent hang.

No production code changed.

Test plan

• dotnet build --nologo -v q — 0 warnings, 0 errors
• dotnet test src/Netclaw.Actors.Tests --no-build --filter "FullyQualifiedName~SessionBindingContractTests|FullyQualifiedName~ProactiveThread|FullyQualifiedName~ErrorCorrelation|FullyQualifiedName~SlackThreadBackfill" — 181 passed, 0 failed, 6s wall time (no ballooning; the raise only affects failure-path latency)
• dotnet slopwatch analyze — 0 issues
• ./scripts/Add-FileHeaders.ps1 -Verify — all files have headers