M11: SQL, remaining IO, and 1.0

Tracking issue for **M11**. Specification: [`06-pandas-parity.md`](../blob/main/docs/specs/06-pandas-parity.md), [`09-quality-bar.md`](../blob/main/docs/specs/09-quality-bar.md).

### Scope

- [ ] `firepanda.sql()`, parsing into the **same logical plan** so the optimizer is shared
- [ ] `query()` and `eval()` for pandas parity
- [ ] An ADBC driver
- [ ] The remaining IO formats tagged `(M11)`: SQL, ORC, fixed width, Excel, HTML, XML, clipboard
- [ ] API review and freeze
- [ ] Complete docstring coverage with runnable examples
- [ ] The migration guide, **organized by pandas function name**

### Exit criteria

- [ ] All 22 TPC-H queries run correctly at SF10, with published timings against DuckDB, Polars and MojoFrame
- [ ] The public API is frozen and every exported symbol is documented
- [ ] Every checkbox in `06-pandas-parity.md` outside the explicitly post-1.0 sections is ticked

### On the migration guide

Somebody arriving at this library is holding a pandas program. The guide is a lookup table from what they have to what they write, and every divergence appears **on the row where it bites** rather than in a general caveats section nobody reads.

### Depends on

M7. Ideally M8 through M10, but 1.0 can ship without M10.

This is a broad M11 tracking issue rather than a self-contained task. Start with docs/specs/06-pandas-parity.md and docs/specs/09-quality-bar.md, then review the M7 dependency and the listed SQL, IO, API, documentation, migration-guide, and TPC-H work. Done means the stated exit criteria are met, including the required parity checklist, API freeze, documentation, and benchmarks.

Written by the indexing model from the issue text.