fix(api): handle DoesNotExist in audit log signals and webhooks

[wavebyrd]

Summary

Fixes #6879

When a FeatureState or FeatureSegment is deleted before the async create_audit_log_from_historical_record task runs, three locations crash with DoesNotExist errors (reported via Sentry as FLAGSMITH-API-5JA, 5JB, 5JD).

This handles DoesNotExist gracefully in all three locations:

• audit/signals.py - send_feature_flag_went_live_signal: catches FeatureState.DoesNotExist when accessing audited_instance.feature_state and skips the signal
• integrations/grafana/mappers.py - _get_instance_tags_from_audit_log_record: catches ObjectDoesNotExist and returns empty tags (annotation is still created, just without feature/segment tags)
• features/tasks.py - _get_feature_state_webhook_data: catches ObjectDoesNotExist when accessing feature_state.feature_segment and falls back to None

Changes

File	Change
api/audit/signals.py	Wrap FK access in try/except, log and skip
api/integrations/grafana/mappers.py	Wrap tag-building in try/except, return []
api/features/tasks.py	Extract FK access with try/except, fall back to None
api/tests/unit/audit/test_unit_audit_signals.py	Test: deleted FeatureState skips signal
api/tests/unit/integrations/grafana/test_mappers.py	Test: deleted FeatureState returns no instance tags
api/tests/unit/features/test_unit_features_tasks.py	Test: deleted FeatureSegment still returns webhook data

Test plan

• test_send_feature_flag_went_live_signal__deleted_feature_state__skips_signal
• test_map_audit_log_record_to_grafana_annotation__deleted_feature_state__returns_no_instance_tags
• test_get_feature_state_webhook_data__deleted_feature_segment__returns_data
• CI passes existing test suite

[vercel]

Someone is attempting to deploy a commit to the Flagsmith Team on Vercel.

A member of the Team first needs to authorize it.

[matthewelwell]

Hi @wavebyrd , similar to your other PR here, I think you've made a mistake in the merge or rebase as it's showing 3000+ files changed.

[wavebyrd]

Fixed the rebase - now showing only the single commit for this fix. Apologies for the messy history!

[emyller]

As I understand these changes, they will eliminate the error by suppressing it.

Similarly to this other pull request, the solution could be suppressing it, but it could also be avoiding it.

Either way, I think this needs deeper investigation, as did the pull request linked above, in order to make sure we're fixing the problem correctly, and not hiding others silently.

@wavebyrd Again, thanks for your contribution. Allow us time to investigate what needs to be done here, or feel free to take another stab with this review in mind — though I think it needs effort to reproduce and investigate the user journey in detail.