fix: accept invalid headers with a space

[ksw2000]

Fix #1917

Make behavior consistent with net/http by allowing header keys and trailers containing spaces without canonicalizing them.

Change the behavior of headerScanner.next(). When parsing headers, we do not normalize the header key.
In addition, before normalizing the key, check whether the key contains a space. If it does, disable normalization forcibly.

[ksw2000]

I think the fuzzy test run failure is independent of my changes since the previous commit passed. The latest commit is a refactor to fix linter problems. In addition, the failed test - fuzzResponseReadLimitBody is irrelevant to the header parser.

[erikdubbelboer]

No, what the fuzzer found is caused by your changes. Here is the test case from the fuzzer that passes on master but panics on your branch:

func TestReadLimitBodyExample(t *testing.T) {
	body := []byte{0xa1, 0xfd, 0xdf, 0x77, 0x20, 0x35, 0xa, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x66, 0x65, 0x72, 0x2d, 0x45, 0x6e, 0x63, 0x6f, 0x64, 0x69, 0x6e, 0x67, 0x3a, 0xb0, 0xa, 0xa, 0x30, 0x52, 0xd, 0xa, 0x3a, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x89, 0x3a, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x89, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0xa, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0xa, 0x20, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0x20, 0x3a, 0x3a, 0xa, 0xa, 0x20, 0x20, 0xa, 0x20, 0xa, 0x20, 0x3a, 0x20, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x20, 0x3a, 0x20, 0xa, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x2e}

	res := AcquireResponse()
	defer ReleaseResponse(res)

	_ = res.ReadLimitBody(bufio.NewReader(bytes.NewReader(body)), 24922)
}

[ksw2000]

This bug is caused by two issues:

1. When refactoring parseTrailer, the usage of continue jumped to the wrong loop, and there was also a logical error in using disableNormalizing. This is not the main problem.
2. In the original implementation, if the key in the trailer was " \n", the program would return an error because of the space. However, after the modification, spaces were allowed, but since the initial logic did not block "\n", the modified program would mistakenly allow this error. In the new implementation, parseTrailer follows the approach of parseHeaders by calling validHeaderFieldByte to check for unexpected characters.

I am trying to minimize the test you provided and write it back as a unit test.

[erikdubbelboer]

Thanks!