[Bug] reject non-canonical git change IDs #796

Description

ParseChangeID accepts Git change IDs that are not in their canonical form.

The parser uses net/url, which can normalize parts of the URI or drop them entirely. As a result, different input strings can be parsed into the same ChangeID.

For example, inputs with an uppercase scheme, userinfo, query parameters, fragments, or non-canonical percent-encoding are currently accepted even though they don't round-trip to the original string.

The parser also accepts empty segments in the repository path.

Steps to Reproduce

1. Call ParseChangeID with a non-canonical Git change ID, for example:
   GIT://git.example.com/uber/monorepo/refs%2Fheads%2Fmain/<sha>
2. Observe that parsing succeeds.
3. Call String() on the returned ChangeID.
4. The resulting string is different from the original input.

The same issue can be reproduced with userinfo, query parameters, fragments, lowercase percent-encoding, unnecessary percent-encoding, and empty repository path segments.

Expected Behavior

ParseChangeID should reject any input that is not in canonical form.

A valid change ID should round-trip exactly:

ParseChangeID(raw).String() == raw

Actual Behavior

Non-canonical inputs are accepted and net/url normalizes or drops parts of them during parsing.

For example, an uppercase GIT:// scheme is parsed as git://, and query parameters or fragments are not included in the resulting ChangeID.

Environment

• Go version: go1.27.1
• OS: Windows
• Bazel version: 8.4.1

Logs / Screenshots

Not applicable.

Additional Context

This can result in multiple different strings representing the same change ID.

The parser should enforce a single canonical representation so that change IDs are unambiguous.