fix: percent-encode skill file paths in download_skill - #5827

Closes #5826

Description

download_skill() built skill://{skill_name}/{file_info.path} from the server's raw manifest paths without percent-encoding. Because the server unquotes path parameters once, a file named 100%20score.md was fetched as 100 score.md (wrong bytes) and a file named notes#1.md was truncated at the URI fragment (failure or a sibling's content). The server already publishes percent-encoded URIs in SkillProvider._list_resources; the client now requests the same form.

The download URI is built with quote(file_info.path, safe='/'), preserving nested paths. The skill-name segment is deliberately left raw: the server emits and matches it without quoting (SkillProvider._list_resources, _get_resource, _get_resource_template), so encoding it would regress names containing legal sub-delimiters or a literal percent (e.g. pdf!, a&b, 100%20score). A parameterized regression test pins those names.

Contribution type

• Bug fix (simple, well-scoped fix for a clearly broken behavior)
• Documentation improvement
• Enhancement

Bug fix verification

Test path: tests/utilities/test_skills.py::TestDownloadSkillUriEncoding (4 tests).

Before the fix (current main adf90b6):

AssertionError: assert 'DECODED SPACE' == 'LITERAL PERCENT'     # 100%20score.md fetched the decoded sibling
AssertionError: assert 'SIBLING WITHOUT FRAGMENT' == 'HASH FILE'  # notes#1.md fetched notes
FileNotFoundError: ...\team\pdf\100%20#1.md                     # fragment-truncated request
3 failed, 5 passed

After the fix: all eight pass, and independent repro scripts download the correct bytes for percent-triplet, hash-shadow, hash-only, nested-space, unicode, and ordinary nested paths. The parameterized skill-name tests pass on current main as well — they guard against re-introducing name encoding (an intermediate revision quoted the name and failed them).

Validation

• uv run --frozen pytest tests/utilities/test_skills.py -q -> 34 passed
• uv run --frozen pytest tests/server/providers/test_skills_provider.py tests/server/providers/test_skills_vendor_providers.py tests/server/providers/test_skill_main_files.py -q -> 120 passed
• uv run --frozen ruff check and ruff format --check on the changed files -> clean

AI assistance

Tool(s) used: Claude Code
How you used it: investigated from a verified reproduction, drafted the fix and regression tests.

Checklist

• This PR addresses an existing issue (or fixes a self-evident bug)
• I have read CONTRIBUTING.md
• I have added tests that cover my changes
• I have run uv run prek run --all-files and all checks pass
• I have self-reviewed my changes
• If I used an LLM, it followed the repo's contributing conventions (not generic output)