fix: preserve tail text in Selector.get_all_text

[tsubasakong]

Summary

• preserve tail text nodes when collecting recursive text content
• keep text order stable across nested children
• add regression coverage for interleaved child/tail text and ignored-tag tails

Fixes #167

Validation

• source .venv/bin/activate && pytest tests/parser/test_parser_advanced.py -q
• source .venv/bin/activate && pytest tests/parser/test_general.py -q -k all_text
• git diff --check

[D4Vinci]

A duplicate of #168, and this PR is against the contribution rules.

Also, PR #168 is the best approach because:

1. Correctness: The ancestor-walk approach for ignored tags is the only one that handles arbitrarily nested ignored elements correctly. This PR still recurses into ignored elements, potentially leaking their content.
2. Performance: Using a pre-compiled XPath(".//text()") at the module level is consistent with the existing codebase patterns (_find_all_elements, _find_all_elements_with_spaces are already pre-compiled the same way).
3. Simplification: It correctly removes the unnecessary _find_all_elements expansion from ignored_elements — since it walks ancestors, it only needs the tag elements themselves in the set, which is simpler and faster.