Preserve URI parse error info in error source chain

[veeceey]

Fixes #3043

When hyper parsed an HTTP request with an invalid URI, it would create a Parse(Uri) error but discard the original http::uri::InvalidUri error. This meant error.source() returned None, making it impossible to get the actual reason the URI was invalid.

Changed Parse::Uri to carry an Option<Cause> so the original InvalidUri or InvalidUriParts error flows through to Error::source(). The error message and Display output are unchanged - the improvement is only visible through the Error::source() chain.

Added a test that verifies error.source() returns the original http::uri::InvalidUri:

running 86 tests
...
test error::tests::uri_error_preserves_source ... ok
...
test result: ok. 86 passed; 0 failed; 0 ignored

[veeceey]

hey, just checking in — is there anything else needed from my side?

[seanmonstar]

I think this feels a bit unexpected. I'd expect all these variants to stay thin without allocations, and the Cause would only ever be used in with().

[veeceey]

Yeah that's a fair point — adding the Option to the Parse variant bloats the enum for all variants even though it's only ever Some during the brief From conversion. I'll rework it so Parse::Uri stays thin and instead thread the cause through a separate path into Error::with(). Give me a bit to push the updated approach.

[veeceey]

Dug into this more and it's trickier than I initially thought. The issue is that parse() returns ParseResult<T> = Result<_, Parse>, not Result<_, Error>. So the URI cause needs to survive through Parse before getting converted to Error at the boundary in io.rs.

Alternatives I looked at:

1. Change ParseResult to use Error instead of Parse — lets us skip the intermediary entirely, but it touches the Http1Transaction trait + every parse call site. Pretty invasive.
2. Handle URI parsing manually (replace ? with explicit map_err) — still can't return an Error from parse().
3. Current approach: Option<Cause> on Parse::Uri — adds ~16 bytes to the Parse enum. The Some only exists briefly between From<InvalidUri> and From<Parse> for Error where it gets moved into .with().

If the enum size bump is a dealbreaker, option 1 is probably the right long-term fix but it's a bigger refactor. Happy to go whichever direction you prefer — or close this if you'd rather tackle it as part of a broader error rework.