v0.2.10

Bug fixes from a candidate-test-run pass

• havn --version now reads from package metadata, so the reported version always matches the installed wheel. Previously hardcoded as 0.2.7.
• Script-runner row-count summary tolerates comma and underscore thousand-separators. A script printing "loaded 2,616,838 rows" no longer reports "845 rows" in the build log.
• CLI forces UTF-8 on stdout/stderr on Windows so non-ASCII data (Norwegian, German, Asian scripts, emoji) renders correctly. Previously characters like `ø` and `å` came out as `?`.
• Parallel transform orchestrator pre-creates target schemas on the main connection before dispatching workers. Fixes "Catalog write-write conflict on create with bronze" race on a fresh `havn init` warehouse where multiple tier-1 models all targeted the bronze schema.

Tests

• New regression test for the row-count comma case.
• New parallel test that builds 4 independent bronze models on a fresh warehouse with parallel=True. Verified to fail without the schema-precreate fix.