[circt-lsp-server] Make time source injectable

[Scheremo]

Introduce an injectable clock and test-only wait shim to remove real sleeps
from PendingChanges logic in tests, while preserving production behavior.

PendingChangesMap Changes

• Add an injectable time source:
  • using SteadyClock = std::chrono::steady_clock;
  • using NowFn = std::function<SteadyClock::time_point()>;
  • Store NowFn nowFn and a bool useManualClock in PendingChangesMap.
• Replace direct calls to std::chrono::steady_clock::now() with nowFn().
• Add waitForMinMs(uint64_t ms, SteadyClock::time_point start):
  • Prod path: std::this_thread::sleep_for(ms).
  • Test path (useManualClock): busy-wait with std::this_thread::yield() until nowFn() reaches target.
• New ctor overload for tests:
  • PendingChangesMap(unsigned maxThreads, NowFn now) -> enables manual clock mode.
• Dtor now calls abort() to ensure all pending work is cleared on teardown.
• Replace sleep in debounce worker:
  • sleep_for(ms) -> waitForMinMs(ms, scheduleTime).

PendingChangesMap Unit Test Changes

• Add ManualClock with manual advancement (advanceMs) and now() accessor.
• Utility advanceTestTime(clock, ms) that advances test time and yields.
• CallbackCapture::waitFor() now uses a tight fixed 1ms window (no per-call timeout arg).
• Refactor tests to:
  • Construct PendingChangesMap with manual clock (PendingChangesMap(2, [&]{ return clock.now(); })).
  • Advance time deterministically instead of sleeping.
  • Validate debounce min/obsolete/max-cap behaviors without flakiness.
• Adjust debounce cap in MaxCapForcesFlushDuringContinuousTyping to 100ms to tighten assertions.
• Sprinkle end-of-test large time advances to flush any queued workers.

[uenoku]

Cool, thank you for improving this! We need to make sure not to use raw now()/sleep_for() but that sounds reasonable

[Scheremo]

Cool, thank you for improving this! We need to make sure not to use raw now()/sleep_for() but that sounds reasonable

Thanks for spotting the issue and fixing it on short notice!
My current approach is to centralize time control in the unit test, so each thread "runs on unit test time". This should remove any race conditions on time alone. I think the draft is almost complete, so please feel free to let me know if you have any feedback 😄

[uenoku]

LGTM, thank you for improving this! Maybe we might want to check abort is called in deconstructor of PendingChangesMap (RAII also works, but I personally prefer requiring explicit abort ).

[Scheremo]

LGTM, thank you for improving this! Maybe we might want to check abort is called in deconstructor of PendingChangesMap (RAII also works, but I personally prefer requiring explicit abort ).

That's a good point; I added a destructor that calls abort. This should help avoid accidental thread orphaning, and if there are no threads to wait on, this should also not cost anything if abort has been called prior to destruction.