[kvlonge]

Why

fensu 0.27 enforces small tuples (FPRSS012), crate layout (FPRSR311) and Python dead-code detection. Upgrading keeps the structure checks current and removes code nothing reaches.

Changes

• Bump fensu to 0.27.

• Replace every tuple over three elements with a named struct. Rows that cross into Python use a shared positional_record!/positional_fields! helper, so Python sends and receives the same tuples as before.

• Move the Snowflake function list under sqlbuild-analysis/src; allow pyproject.toml in crate roots.

• Enable [targets.python.dead_code] and remove dead code (~350 lines):

  • adapter constants now owned by Rust (contract, BigQuery, Snowflake, ClickHouse, SQLite)
  • two executor models and two Dagster helpers, plus their constants
  • tooling: the random DAG project, CaptureFile (its text helper is kept), policy constants
  • six empty packages

• Roots for code reached by mechanisms fensu cannot see:

  • _native_host targets and built-in adapter classes
  • python -m script entry points and fensu rule modules
  • tooling run only by tests
  • Dagster/rivers exhaustiveness sets
  • protocols used in _native.pyi

  A test keeps the host roots in sync with the host table.

• Raise the XSB073 budgets: a named struct's Python fields are now counted, where tuple positions were not.

Verification

• fensu check: 0 faults (164 before, on 0.27).
• make compiler-differential passes; goldens unchanged.
• Focused pytest and cargo test on the touched crates pass.
• make test: 11750 passed. One wall-time budget test failed under load and passes on its own.
• make check-ci passes via verify; hygiene scan is clean.

[kvlonge]

Closing in favour of one combined PR (H2 + fensu 0.27 + SharedMergeTree on top of the merged deadlock fix) so these share a single CI run; the branch is being fixed for an interaction between H2's broken-pipe parity tests and the merged changes.