feat(source-hubspot): capture CRM deletions as in-stream tombstones

[vklimontovich]

What

Adds deletion capture to source-hubspot. HubSpot CRM streams are read incrementally through
POST /crm/v3/objects/{entity}/search, but that endpoint never returns archived (deleted) records —
so once an object is deleted in HubSpot it silently stops updating and the stale row lives forever;
deletions are invisible on Incremental Append. (See airbytehq#47198 and airbytehq#40595.)

This adds a HubspotDeletionRetriever custom component. For each opted-in CRM search stream it runs the
normal live search pass plus a second pass against GET /crm/v3/objects/{entity}?archived=true, and
emits each archived object back into the same stream as a CDC tombstone:

• same primary key id
• the full flattened properties_* set (so a destination merge does not NULL-overwrite columns)
• _ab_cdc_deleted_at = archivedAt (fallback createdAt)
• updatedAt = archivedAt (so the tombstone sorts after the live row for PK dedup)

Emitted exactly once per sync (lock-guarded — the concurrent path shares one retriever across
time-slices) and best-effort: if the archived endpoint is unavailable for a scope it logs a warning and
continues, never failing the live sync.

Covered streams

companies, contacts, deals, leads, engagements_calls, engagements_emails,
engagements_notes, engagements_tasks. The existing deals_archived stream is left unchanged.

Deferred (and why)

tickets, engagements_meetings, deal_splits, goals — HubSpot rejects archived paging for these
("Paging through deleted objects is not yet supported" / 403). products, line_items — follow-up.

Design note for maintainers

In-stream tombstones need no new streams or catalog changes and work with existing incremental+dedup
destinations. The repo already has a precedent for the separate *_archived stream approach (e.g.
deals_archived), which is what airbytehq#40595 literally asks for. If you'd prefer that direction instead of/in
addition to CDC tombstones, happy to adjust — flagging before it lands.

Testing

• New unit_tests/test_hubspot_deletion_retriever.py: an archived record becomes a tombstone with
  _ab_cdc_deleted_at == updatedAt == archivedAt, the same id, and the full flattened properties_*.
• Manifest parses and spec emits; all 8 wired streams resolve HubspotDeletionRetriever.
• Validated against a live HubSpot account: companies/deals/contacts tombstones with correct markers,
  no duplicates, full properties.

Version

Bumps dockerImageTag 6.7.0 → 6.8.0 and adds a changelog entry.