perf: consider in-place en- and decryption

[mxinden]

We currently re-allocate when en- and decrypting:

neqo/neqo-transport/src/crypto.rs

Lines 664 to 665 in f3d0191

	let size = body.len() + MAX_AUTH_TAG;
	let mut out = vec![0; size];

neqo/neqo-transport/src/crypto.rs

Line 689 in f3d0191

let mut out = vec![0; body.len()];

Now that we tackled various low-hanging fruits around memory allocation (e.g. #2184), memory allocations when en-/decrypting become more prominent. See e.g. the following recent Firefox profile doing a large file transfer:

https://profiler.firefox.com/public/y1ck0xgt3102jyqjjp93ggda99dt8bz2yg5c7k8/flame-graph/?globalTrackOrder=01&thread=1&transforms=ff-72~ff-82~ff-83&v=10

[mansf-osk]

I took a brief look just now and would be interested to dig further when I get back from vacation in February. Maybe we can have a chat then 🙂

[larseggert]

I actually started on this a few days ago. See PR.