fix: nest role type in update_opportunity_roles payload (unbreak set-role)

[jschfflr]

The bug

update_opportunity_roles (backing deals set-role) sent the role type flat on each entry:

{"opportunity_id": "…", "roles": [{"contact_id": "…", "opportunity_contact_role_type_id": "…", "is_primary": true}]}

Apollo 422'd every call with undefined method 'map' for nil — its handler calls .map on each entry's role field, which was absent. The command never worked against the live API (it passed only because tests mock the client).

Root cause

Captured the app UI's own update_roles call: each entry nests the role type under a role array:

{"contact_id": "…", "is_primary": true, "role": [{"opportunity_contact_role_type_id": "…", "is_primary": true}]}

Fix

update_opportunity_roles now reshapes the flat RoleAssignment entries into that nested wire format (role type omitted from the nested object when the caller didn't supply one). The public RoleAssignment interface is unchanged — callers still pass flat entries.

Testing

• Updated test_update_opportunity_roles to assert the nested body (both with- and without-role-type entries).
• Live-verified end-to-end on a real deal: linked a contact as primary, then flipped its role type — both persisted (num_contacts=1, correct role type on GET).
• ruff ✅ · pyright ✅ · 181 tests ✅

Note: no CLI change needed (deals set-role already sends flat entries). No version bump (left for release time).

🤖 Generated with Claude Code

[peqy]

Fix looks correct and matches the wire format Apollo expects; test coverage was updated appropriately.

Found 1 issue (0 critical, 1 improvement).

[peqy]

Improvement: is_primary = bool(entry.get("is_primary")) will treat any truthy non-bool (e.g. the string 'false') as True, and it also forces an omitted value to False.

Fix: consider is_primary = entry.get("is_primary", False) (and optionally isinstance(is_primary, bool) / raise) so you don’t accidentally flip primaries due to Python truthiness.

[peqy]

👍 Update looks good — switching to entry.get("is_primary", False) avoids the truthiness pitfall I mentioned. Feel free to mark this thread resolved.