fix: multichunk reads

[SirBrucey]

When EPP responses arrived in multiple chunks, the read counter would reset to its previous value instead of accumulating.
This caused the code to think it hadn't read as many bytes as it actually had, leading to "Unexpected EOF while reading" errors.

2026-02-05T12:02:05.416810Z DEBUG: instant_epp::connection: Expected response length: 2539
2026-02-05T12:02:05.417597Z DEBUG: instant_epp::connection: Read 1172 bytes, 1428 out of 2539 done
2026-02-05T12:02:05.417637Z DEBUG: instant_epp::connection: Read 1111 bytes, 1367 out of 2539 done

The connection state machine's pattern destructuring creates copies of state fields instead of mutable references into the struct. When these were modified (e.g., incrementing read counter), the modifications were lost because the original state was returned.

Example:

RequestState::Reading { mut read, buf, expected } => {  // copies read
  ...
  read += filled.len();  // Modifies LOCAL copy
  ...
  Transition::Next(state)  // Returns ORIGINAL state with OLD read value
}

Fix: Construct new state objects with updated values:

RequestState::Reading { read, buf, expected } => {
  let new_read = *read + filled.len();  // Calculate new value

  // Return NEW state with updated counter
  Transition::Next(RequestState::Reading {
      read: new_read,
      buf: mem::take(buf),
      expected: *expected,
  })
}

[djc]

Nice, thanks for fixing this upstream and adding test coverage!

Would be curious to hear what/where you're using it.

[djc]

I'm not a fan of adding this new dependency with unfamiliar attribute macro stuff. Can you just write a few #[test] functions with maybe a simple helper function?

[djc]

Nit: I'd prefer to just shadow the old binding, keeping it as start (here and below).