✨ feat(query): add XPath 2.0 string functions

[gaborbernat]

Adds the string convenience subset of XPath 2.0 to the native engine, so expressions ported from elementpath, lxml, or htmlquery that reach for these functions run without registration:

• ends-with(a, b) — suffix test, the mirror of the existing starts-with.
• string-join(seq, sep) — the string-values of a node-set joined by a separator (the XPath-1.0-engine reading, where a node-set is the sequence).
• lower-case(s) / upper-case(s) — Unicode case mapping, delegated to CPython's str.lower()/str.upper() the same way the re: functions call into CPython.
• matches(input, pattern[, flags]) — regex test, sharing the EXSLT re:test pipeline.
• replace(input, pattern, repl[, flags]) — regex replace of every match, with $N group references rewritten into the re module's template form.

All dispatch in the compiled-C function table alongside the 1.0 core and the EXSLT namespaces; the full XPath 2.0 sequence, type, and FLWOR machinery stays out of scope. reverse() is left out because the engine keeps every node-set in document order and re-sorts after each step, so a reversed set could not survive a path.

Competitor design was drawn from elementpath (arg orders, flag letters, the $N→group rewrite), antchfx/xpath (node-set string-join, 2-arg ends-with), and the W3C XPath/XQuery F&O spec for exact semantics.

closes #542

[codspeed-hq]

Merging this PR will not alter performance

✅ 62 untouched benchmarks
⏩ 18 skipped benchmarks¹

---

_{Comparing gaborbernat:feat/542-xpath2-functions (ebb3af2) with main (053eeb2)}

Footnotes

1. 18 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. ↩