Add MSSQL bulk load via BCP

[martinv13]

Summary

• Adds a bulk_insert() override to MSSQLDialect that uses the bcp
  utility for large batches (≥ 1000 rows), falling back to
  fast_executemany for smaller batches or when bcp is not on PATH
• Auto-detects bcp at DataModel creation time via shutil.which();
  no configuration needed when available
• Pass use_bcp=False to DataModel to opt out entirely
• Threads use_bcp through DataModel.__init__() → get_dialect() →
  dialect constructor via **kwargs, so other dialects are unaffected

Value formatting (BCP character mode, tab-separated)

Python type	Written as	Notes
None	empty field	-k flag maps to NULL
bool	0 / 1	BIT columns
bytes	{hex}	VARBINARY hex literal
datetime	str(v)	ISO format SQL Server accepts
str	str(v), tab→space	tab is the field delimiter

Notes

• BCP does not participate in the caller's SQLAlchemy transaction —
  acceptable since insertions target ephemeral staging tables that are
  dropped after each load
• MSSQL CI workflow updated to install mssql-tools18 (provides bcp)
  and expose it on PATH

Test plan

• tests/test_bulk_insert_mssql.py — 10 tests: 4 covering the
  fast_executemany path (< 1000 rows: basic, numeric, boolean, scalar
  default) and 5 covering the BCP path (basic, numeric, boolean, binary,
  scalar default), plus 1 fallback test that forces bcp_path = None to
  verify graceful degradation
• BCP tests skip automatically when bcp is not on PATH
• All non-DB tests continue to pass (pytest -m "not dbtest")

🤖 Generated with Claude Code