Promote high-value extra_json fields to explicit columns

[dmwyatt]

Context

From the API data gaps investigation (see #177 for background).

The get-documents-v2 API returns 42 fields per document. Our Document struct explicitly models 20 of them. The remaining 22 are captured by #[serde(flatten)] pub extra: HashMap<String, Value> and serialized to the extra_json column. This means these fields are not independently queryable.

Fields worth promoting

Potentially useful for queries:

Field	Type	Notes
selected_template	string/null	Template ID used for AI generation; useful for filtering
valid_meeting	bool/null	Whether Granola considers this a real meeting
transcript_deleted_at	datetime/null	When transcript was deleted (retention policy)
was_trashed	bool/null	Whether document was previously trashed
is_primary_event_note	bool	Relationship to calendar event
attachments	array/null	Meeting attachments
audio_file_handle	string/null	Reference to audio recording
overview	string/null	Currently null for all docs; may be populated in future
chapters	array/null	Currently null for all docs; may be populated in future

Lower priority (integration/metadata):

Field	Type	Notes
hubspot_note_url	string/null	CRM integration
affinity_note_id	string/null	CRM integration
attio_shared_at	datetime/null	CRM integration
external_transcription_id	string/null	External transcription reference
cloned_from	uuid/null	Document lineage
meeting_end_count	number	Meeting lifecycle metadata

Proposal

Add explicit columns for the high-value fields via a migration. Start with valid_meeting, selected_template, and transcript_deleted_at as they're most likely to be useful for filtering and queries.