fix(mysql): return error instead of panic on truncated OK packet

[cvzx]

Does your PR solve an issue?

fixes #4139

Is this a breaking change?

No. This changes a panic into a returned Err(Protocol(...)), which callers already handle. The connection pool will gracefully discard the bad connection instead of crashing the tokio worker thread.

[abonander]

I don't want to move the goalposts too much, but it seems like these fields should at least be covered by the length check too.

In the long run we should probably ban use of bytes::Buf methods and instead write our own that return Result.

[cvzx]

Checking get_uint_lenenc calls is a bit trickier since the method advances a variable number of bytes depending on the prefix. I've changed its signature to return Result and added remaining-bytes checks inside the method, propagating with ? at all call sites.

If this approach is good, I could follow up with a separate PR that wraps the other panicking bytes::Buf methods in the same way.

[abonander]

I was mostly thinking that we should just check that the packet is nonempty, but this is a great improvement, thanks!

If this approach is good, I could follow up with a separate PR that wraps the other panicking bytes::Buf methods in the same way.

I think that would be a good idea, there's a lot of shortcuts that we took early on that are coming back to bite us and using things like this that panic instead of returning errors is a big one.