refactor: remove 50 backend and scheduler symbols no production code references (#3433)

[vybe]

Fixes #3433. One of five independent stale-code PRs off dev. Draft: lands after #3440 (the scheduler process-schedule removal), which touches the same scheduler files.

What

• 50 dead symbols removed across backend services, adapters, db modules, utils and the scheduler — every one re-verified to have no production caller on the live tree immediately before deletion (backend, scheduler, MCP server, base image, scripts, compose files, the pinned enterprise submodule, and by-name dispatch through the database facade), plus 5 symbols that became unreferenced as a result (get_active_count, update_schedule, AgentTaskMetrics, _DEFAULT_CHANNEL_TIMEOUT, the _backlog_service global) and the imports and constants only they used.
• Kept, deliberately: distinct_failing_count and registered_columns (their docstrings declare them test/observability accessors); SshService.cleanup_agent_credentials (see Handoffs).
• No lost call site found. The heavily tested _sanitize_filename in adapters/message_router.py was a byte-identical duplicate of the live upload_service.sanitize_filename; its 8 tests now run against the live function.
• Tests re-pointed, not dropped, where they covered live behaviour: the freeze-policy asserts now call sync_freeze_reason(...) is not None; the bulk heartbeat test gained the payload asserts the deleted single-agent test carried; two new sync-loop tests prove a disabled schedule is never registered and is dropped when disabled between ticks (previously untested).
• Scheduler test fixture: the same two one-line commits as refactor(scheduler): remove the process-schedule remnant (#3432) #3440 (chain_depth, claim_token).
• Docs: the memory docs that named removed symbols are corrected, including ssh-access.md (it claimed the SSH cleanup runs at agent stop/delete; nothing calls it) and the snippets that described a scheduler client module removed in v0.7.0.

No schema, migration, route or MCP-tool change.

Rulings carried (orchestrator, on the operator's behalf — plan file)

• Planner's classification accepted: 50 remove, 2 keep.
• reload_schedules and the other push-style scheduler methods are removed here; no landing order is imposed by this branch, but see Before merge.
• The memory-doc lines are edited in this PR rather than left to docs: memory-doc drift — architecture.md lists 2 deleted modules, 9 flow docs describe removed surfaces, 26 point at moved paths, index hygiene #2494.
• cleanup_agent_credentials is not deleted; it is reported.
• The fixture fix rides as the identical commits carried by refactor(scheduler): remove the process-schedule remnant (#3432) #3440 instead of a separate precursor PR.

Review + security

/review (reading review, report-only): MERGEABLE, 0 critical. Independent per-name sweep at the branch tip: zero live references, nothing exported through __all__ or a package __init__. Each security-relevant removal has live code performing the same check today (Slack signature verification in the webhook handler, validate_password_strength, inline webhook-secret decryption, the allow-list 403). /cso diff pass: no findings, net surface reduction. Fixed after review: the sync-loop disabled-schedule tests and the two doc corrections.

Tests

• tests/scheduler_tests: 237 passed, 0 failed.
• All 20 unit files this branch edits, shuffled (seed 12345), one per process: 659 passed, 32 skipped, 0 failed.
• New sync-loop tests are mutation-proven (each fails when its branch is broken).
• Import smoke: backend main (674 routes) and scheduler.main import; compileall clean.
• git merge-tree against dev (21b4a11): clean.
• Not run here: the full unit island (CI); two edited files under tests/integration/ need a live Redis and were collect-checked only.

Before merge

• Order: refactor(scheduler): remove the process-schedule remnant (#3432) #3440 first. This branch conflicts with refactor(scheduler): remove the process-schedule remnant (#3432) #3440 in four files — src/scheduler/service.py, tests/scheduler_tests/test_1296_reminders.py, tests/scheduler_tests/test_sync_loop.py, docs/memory/feature-flows/scheduler-service.md. In each, both branches delete or append adjacent blocks; the resolution is to keep both sides' changes. Whichever lands second resolves them.
• fix(tests): repair tests and fixtures drifted behind merged changes (#3442, #3272) #3468 (test-drift repair) adds the same two columns to tests/scheduler_tests/conftest.py in one hunk and conflicts with the two fixture commits here. If fix(tests): repair tests and fixtures drifted behind merged changes (#3442, #3272) #3468 lands first, take its version of that hunk.
• Run the two edited integration test files against a live Redis.

Handoffs

• SshService.cleanup_agent_credentials has no caller: deleting or stopping an agent never purges its SSH key metadata. Pre-existing; to be filed as its own bug. Its docstring in ssh_service.py still says it is called on stop/delete.
• The channel_timeout_seconds setting has had no reader since fix: Slack message router ignores per-agent execution timeout (TIMEOUT-001) #221.
• A dead-code ratchet in CI was considered and not built; see refactor: remove dead code flagged by vulture (unused imports/variables) #1032.

🤖 Generated with Claude Code

[trinity-ability]

Reviewed: per-name sweep finds no live reference on dev, enterprise submodule or by-name facade dispatch; 25,112 unit tests pass (the 1 failure, test_1160 concurrent boot, also fails on clean dev). Conflicts with #3440 resolved as union of deletions; scheduler suite 235 passed on the merged tree. The regression-diff failure is test_2163's 0.02s-vs-0.05s timing race in portal briefings, a file this PR does not touch; passes locally.