docs: add long description and metadata for the PyPI page

[d-v-b]

🤖 AI text below 🤖

Closes #26.

Problem

https://pypi.org/project/cast-value-rs/ shows no information about the package. python/pyproject.toml set description (the one-line summary shown under the package name) but no readme, so PyPI had no long description to render as the page body.

Change

• Add python/README.md as the long description: what the codec does, why the implementation is in Rust, installation, the four usage patterns from the docs (basic cast, dtype spellings, scalar maps, pre-allocated output), a supported types/modes table, and links to the docs site and licenses.
• Set readme = "README.md" in python/pyproject.toml.
• Add keywords, a [project.urls] table (Homepage / Documentation / Repository / Issues) to populate the PyPI sidebar, and license, audience, topic, and Typing :: Typed classifiers.

The README lives under python/ rather than referencing the repo-root one, so it ships in both the sdist and the wheels — the CD workflow runs maturin with working-directory: python, and a ../README.md path would not survive packaging.

Verification

Built both artifacts locally with maturin and confirmed the metadata lands:

• sdist and wheel METADATA both carry Description-Content-Type: text/markdown, the full README body, Keywords, all four Project-URL entries, and the new classifiers.
• twine check passes on both (this is the check that catches a long description PyPI would refuse to render).
• README.md is present in the sdist.

Every claim in the README was executed against the built wheel rather than trusted from the existing docs: all four code snippets produce the exact outputs shown, all 11 dtype names and 5 rounding modes are accepted, clamp/wrap work, and omitting out_of_range_mode raises as documented.

Note that the PyPI page will only update on the next release.

[codspeed-hq]

Merging this PR will not alter performance

✅ 51 untouched benchmarks
⏩ 21 skipped benchmarks¹

---

_{Comparing docs/pypi-description (22a7af8) with main (7f46619)}

Footnotes

1. 21 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩