Repository resources fetch the wrong content when a file name contains "%" #3614

Describe the bug

repo:// resource reads build the raw.githubusercontent.com URL from the decoded file path, so a file name containing % fetches the wrong content. For docs/100%.txt, resources/read on repo://owner/repo/contents/docs/100%25.txt (the URI get_file_contents emits) returns github.com's homepage HTML as the file's text; for a file named %41.txt it returns the contents of A.txt. Spaces, #, ? and unicode are encoded correctly.

Affected version

main @ latest-release-9-g84f423e

Steps to reproduce the behavior

c, _ := raw.NewClient(gh, mustParse("https://raw.githubusercontent.com/"))
c.URLFromOpts(nil, "o", "r", "100%.txt") // "https://raw.githubusercontent.com/"
c.URLFromOpts(nil, "o", "r", "%41.txt")  // ".../o/r/HEAD/%41.txt", served as A.txt

https://raw.githubusercontent.com/ redirects (301) to github.com and comes back as 200 text/html, which the resource handler treats as a successful text read.

Expected vs actual behavior

Expected: the file named 100%.txt / %41.txt is returned. Actual: GitHub's homepage HTML / the contents of A.txt.

Root cause and suggested fix

pkg/raw/raw.go:95 and :113 pass decoded path text to url.URL.JoinPath, which treats its arguments as already escaped: a literal %xx is decoded a second time, and an invalid escape makes JoinPath drop its internal error and keep the base URL. repository_resource.go:192 passes the path already decoded by uritemplate.

url.PathEscape each /-separated segment of owner/repo/ref/path before JoinPath (the .. rejection stays as is). One existing test, benign percent-encoded filename allowed, expects %2ehidden to pass through unescaped, which actually requests .hidden; with the fix it becomes %252ehidden. I have a patch with regression tests (raw URL cases and a resources/read handler test) and can open a PR if the approach sounds right. Related: #3437.