Task: Fix profile binding — single canonical point in * Flags

Unify :profile: binding to a single canonical point: the * Flags section's :PROPERTIES: drawer. Add validation that throws if :profile: is found in the file-level drawer, and revert the manual workaround on 4 reporting entities.

• read_physical_space_overrides reads :profile: from * Flags, not file-level.
• File-level :profile: raises a clear validation error.
• All entity models have :profile: only in * Flags — audit passes.
• Eventing integration tests are generated for entities using any domain profile, without per-entity file-level edits.
• Build passes, drift checks pass, site builds.

1. Audit the two binding points and their differences.
2. Fix read_physical_space_overrides in org_loader.py to read :profile: from * Flags and reject file-level.
3. Revert the file-level :profile: workaround on 4 reporting entities.
4. Verify eventing tests are generated for reporting AND refdata entities.
5. Build, regenerate, run tests.
6. Commit and push.

The codegen has two profile binding points with different behaviour — adding a Physical-space facet override to a profile file only works for entities that declare :profile: in their file-level :PROPERTIES: drawer, which none do:

Every entity model declares :profile: in its * Flags section — that's the existing convention. But read_physical_space_overrides (org_loader.py:183) only reads doc.file_properties.get("profile"), missing every entity's profile.

Fix: read_physical_space_overrides should also find :profile: from the * Flags section (the same place _profile_namespace_defaults reads it), so a single binding point in * Flags drives both feature assignments and facet enablement. Then revert the file-level :profile: additions made to the 4 reporting entity models as a workaround.

Discovered during the reporting-cpp safe drift task in sprint 25. We added a * Physical space table to all 6 domain archetype profiles to enable ores.cpp.eventing-integration-test by default. Entity regeneration produced the test files for reporting entities only because we manually added :profile: to their file-level :PROPERTIES: drawers. Other entities (e.g. all of refdata) silently got nothing — the profile Physical-space table had no effect. The two-binding-point design makes adding a facet to a profile require per-entity edits, defeating the purpose of profiles.

• org_loader.py:183 — read_physical_space_overrides reads file-level only
• Simple lookup profile — example profile with Physical-space table added
• ores.cpp.eventing-integration-test facet — the facet gated by this mechanism
• eventing_integration_test feature — created alongside this work

• Refactor ores.codegen C++ generation — the story where this was discovered

• Fixed read_physical_space_overrides in org_loader.py to read :profile: from * Flags section instead of file-level :PROPERTIES: drawer. File-level :profile: now raises a clear validation error.
• Reverted the file-level :profile: workaround from the 4 reporting entities (concurrency_policy, report_type, report_definition, report_instance).
• Verified: book_status (refdata, simple-lookup) gets _eventing_integration_tests.cpp from its * Flags profile without any per-entity edits.
• 7 files changed: 1 codegen fix, 4 entity reverts, 1 task doc, 1 story update.
• Build 100%, all CI checks pass.