Weird case clause using OTP master with binary data from a NIF

[essen]

Describe the bug
On CI using OTP master I occasionally get the following:

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
rfc9114_SUITE:do_receive_response failed on line 2400
Reason: {case_clause,<<>>}
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

https://github.com/ninenines/cowboy/actions/runs/12768790159/job/35643757234?pr=1658#step:5:1309

But line 2400 is as follow:

	Body = case Rest of
		<<>> ->
			<<>>;
		<<
			0, %% DATA frame.
			DLenEnc:2, DLen:DLenBits,
			Body0:DLen/bytes
		>> ->
			BodyLen = integer_to_binary(byte_size(Body0)),
			Body0
	end,

https://github.com/ninenines/cowboy/blob/master/test/rfc9114_SUITE.erl#L2400-L2410

So I don't think it should give a case clause.

Note however that the binary data is sent from a NIF (quicer) so perhaps that impacts things. However only seems to happen like this on master, and only in this case block (the binary received gets parsed properly earlier, only when looking at the Rest does it lead to a case clause).

To Reproduce
Other than cloning Cowboy and running make ct-rfc9114 COWBOY_QUICER=1 I'm not sure, but even then it might not always happen.

Expected behavior
I don't expect a case clause there.

Affected versions
OTP master.

Additional context
Please provide guidance to better investigate the problem.