fix(db): flush session before accessing deduplication_event.id (fixes #5496)

[asheesh-devops]

Summary

Fixes KeyError: "Deferred loader for attribute 'id' failed to populate correctly" in create_deduplication_event() under high alert load (~1000+ alerts/minute).

Root Cause

In create_deduplication_event(), session.add(deduplication_event) is followed by session.commit(), which expires all ORM attributes. When deduplication_event.id is accessed in the logger after commit, SQLAlchemy triggers a deferred load to re-fetch the ID — this requires a new DB round-trip which fails under high load:

# Current (broken under load):
session.add(deduplication_event)
session.commit()  # Expires all attributes
logger.debug(..., extra={"deduplication_event_id": deduplication_event.id})  # Deferred load fails!

The Fix

Add session.flush() to force the INSERT and populate the auto-generated ID, then capture it in a local variable before commit() expires the ORM object:

# Before (broken):
session.add(deduplication_event)
session.commit()
logger.debug(..., extra={"deduplication_event_id": deduplication_event.id})

# After (fixed):
session.add(deduplication_event)
session.flush()  # Force INSERT, ID now populated
deduplication_event_id = deduplication_event.id  # Capture before expire
session.commit()
logger.debug(..., extra={"deduplication_event_id": deduplication_event_id})

This follows the same session.add() → session.flush() pattern already used in create_incident_from_dict() (line 2420) and _create_application_based_incident() (fixed in #6174).

Changes

• keep/api/core/db.py — add session.flush() and capture ID before commit in create_deduplication_event()

Testing

• Verified the same flush() → capture → commit() pattern is used in create_incident_from_dict() (line 2420-2425)
• The fix is identical in nature to fix(topology): persist incident before adding alerts (fixes #5463) #6174 which was merged for the same class of bug in topology processing
• No behavioral change — only timing of when the ID is read from the ORM object

Fixes #5496 (Issue 2: Deferred loader failure after INSERT)