fix: isolate kanban SQLite board failures

[the24thLetter]

Summary

• classifies Kanban SQLite failures with board-local diagnostics and operator actions
• quarantines corrupt/WAL/disk I/O board failures by DB fingerprint while retrying busy/locked ticks
• exposes embedded dispatch health to the dashboard and verifies gateway logging/no-traceback behavior

Test Plan

• env -u HERMES_KANBAN_REQUIRE_REVIEW_BEFORE_DONE -u HERMES_KANBAN_MERGE_CAPTAIN_PROFILE python -m pytest tests/gateway/test_kanban_dispatch_quarantine.py tests/hermes_cli/test_kanban_db.py::test_connect_exposes_board_diagnostic_for_corrupt_header tests/hermes_cli/test_kanban_db.py::test_connect_wraps_wal_disk_io_error_with_board_diagnostic tests/hermes_cli/test_kanban_db.py::test_dispatch_once_returns_degraded_result_for_release_stale_claims_read_error -q
• python -m py_compile gateway/run.py hermes_cli/kanban_db.py

Note: Running the broader tests/hermes_cli/test_kanban_db.py under the Kanban worker environment inherits HERMES_KANBAN_REQUIRE_REVIEW_BEFORE_DONE/HERMES_KANBAN_MERGE_CAPTAIN_PROFILE and causes unrelated review-gate expectation failures unless those env vars are unset.

[the24thLetter]

Closed: this branch is based on the fork-local integration branch, not upstream main; opening the correct fork-local PR instead.

[alt-glitch]

Superseded by #33929 (same author, same fix — isolate kanban dispatcher SQLite board failures). This was the intermediate attempt.