fix: Reset AbortController before starting generator to prevent infinite abort loop

[TranslateMe]

Problem

Sessions can get stuck in an infinite "Generator aborted" loop, preventing any observation processing.

Root Cause

When a generator exits with wasAborted=true, the cleanup in startGeneratorWithProvider() correctly sets:

• generatorPromise = null ✓
• currentProvider = null ✓

But it does NOT reset the AbortController, leaving signal.aborted = true.

When new observations arrive:

1. ensureGeneratorRunning() sees generatorPromise = null
2. Attempts to start a new generator via startGeneratorWithProvider()
3. New generator immediately detects signal.aborted = true
4. Generator exits with "Generator aborted" message
5. Loop repeats infinitely

Reproduction

1. Start a session and process some observations
2. Abort the generator (e.g., session timeout, crash recovery)
3. Send new observations to the same session
4. Observe infinite "Generator aborted" messages in logs
5. Observations queue grows but never gets processed

Solution

Add a check at the beginning of startGeneratorWithProvider() to detect if the session's AbortController signal is already aborted. If so, create a fresh AbortController instance before proceeding.

if (session.abortController.signal.aborted) {
  logger.debug('SESSION', 'Resetting aborted AbortController before starting generator', {
    sessionId: session.sessionDbId
  });
  session.abortController = new AbortController();
}

This mirrors the existing crash recovery logic (line 192) but applies it universally to all generator starts.

Testing

• Verified fix resolves stuck sessions
• Confirmed new generators start successfully after AbortController reset
• Observation processing resumes normally

Impact

Low risk - only affects sessions that were previously stuck. Normal session flow unchanged.