Normalize one way of checking for frame space

[matheus23]

Currently we check for frame space in a couple of different ways:

https://github.com/n0-computer/quinn/blob/7663451642bffe9077428c6f7e1cb5c3b4f4da97/quinn-proto/src/connection/mod.rs#L5258-L5262

https://github.com/n0-computer/quinn/blob/7663451642bffe9077428c6f7e1cb5c3b4f4da97/quinn-proto/src/connection/mod.rs#L5295-L5299

And sometimes we don't even compare against a constant:

https://github.com/n0-computer/quinn/blob/7663451642bffe9077428c6f7e1cb5c3b4f4da97/quinn-proto/src/connection/streams/state.rs#L450-L451

We should probably use the same way of doing this in all places.

Comparing non-strict (using < and >) also seems to potentially waste another byte of space in the datagram, IIUC. Since if SIZE_BOUND is exactly equal to the frame space remaining, the frame should (in the worst case) fit into the datagram exactly.

---

In a PR review I commented some ideas on how we could refactor this.

[flub]

Possibly a builder.frame_space_for(10) type API instead of .frame_space_remaining() might avoid these variations.

[matheus23]

There are also some other ideas linked above: #295 (comment)