Fix predicate numeric probe erroring on multi-item sequences

[boukeversteegh]

Fixes #147.

Summary

• pop_is_numeric used Value::atomized_option, which raises XPTY0004 on any sequence of length > 1. That turned predicates like a[b] into errors whenever an <a> had more than one <b> child, instead of falling through to the effective-boolean-value path the XPath spec requires.
• The fix returns false for empty or multi-item atomized sequences so the EBV path handles them. Only a single atomized value can be a numeric predicate; anything else is not numeric by definition.

Behavior change

• a[b] (and //lambda[body/block/expression_statement], etc.) now match correctly when the inner path can yield more than one node.
• (1, 2, 3)[(2, 3)] still errors, but now with FORG0006 — the correct spec error for applying EBV to a heterogeneous non-node multi-item sequence. One snapshot updated to reflect this.

Tests

• New regression test_node_predicate_with_multiple_children covering //a[b] against <root><a><b>…</b></a><a><b>…</b><b>…</b></a></root>.
• cargo test --workspace — all green.
• cargo clippy --all-targets --all-features -- -D warnings — clean on Rust 1.95.0.

QT3 conformance gain

Re-running xee-testrunner update vendor/xpath-tests/ shrank vendor/xpath-tests/filters by 5 tests under prod-Predicate, all the same bug class:

K-FilterExpr-38   (1, 2, 3)[(xs:anyURI("..."), xs:anyURI("..."))] — expects FORG0006
K-FilterExpr-40   (similar)
K-FilterExpr-92   (1, 2, 3)[1, 2]                                  — expects FORG0006
K-FilterExpr-93   (1, 2, 3)[1, "a string"]                         — expects FORG0006
K-FilterExpr-94   (1, 2, 3)["a string", 1]                         — expects FORG0006

Before the fix these threw XPTY0004 (wrong error). After the fix they throw FORG0006 (EBV-on-heterogeneous-non-node-sequence), matching the spec. The filter update is included in this PR so conformance tracking stays tight.

Bundled clippy cleanup (happy to split if preferred)

CI runs on the floating stable toolchain and Rust 1.95.0 promoted a few lints to errors. Fixed in a separate commit on this branch:

• interpret.rs: collapsible_match — merged nested if into a guarded match arm.
• library/map.rs (×2): useless_conversion — dropped .into_iter() inside Vec::extend.
• xslt-ast/whitespace.rs: collapsible_match — same pattern.

Purely mechanical; no behavior change. If you'd rather merge this as a standalone PR, say the word and I'll split the commit out.

🤖 Generated with Claude Code