Skip pathless definitions when collecting document symbols

[Eric3-jp]

Importing a collections.namedtuple from another module can make document symbols return an empty result. Jedi reports the imported class in the current document, but its generated members have module_path=None, so processing those members raises TypeError in Path(None) before the remaining local symbols are returned.

Skip definitions without a source path before comparing paths. This preserves the imported class and the rest of the document's symbols while respecting include_import_symbols. The regression uses real files and Jedi, covering both values of all_scopes and include_import_symbols.

Fixes #635.

Validation on Python 3.11.16 / Linux (WSL), with Jedi 0.20.0:

• The four regression cases produced 2 failures and 2 passes before the fix; all four pass afterward.
• Full CI=true pytest -q test/, including the separate /tmp/pyenv environment required by the CI workflow: 206 passed, 9 skipped.
• Ruff 0.9.4 lint and formatting checks pass for pylsp and test.
• JSON schema validation and generated configuration documentation comparison pass.

This contribution was prepared with OpenAI Codex. The behavior and validation above were reproduced locally.