fix(server): prohibit the length headers on successful CONNECT

[ubnt-intrepid]

fixes #1783.

[seanmonstar]

Thanks! Out of curiosity, do you know if the additional branches affect performance much? Likely the ones with least noise are in proto::h1::role at the bottom of the file.

[ubnt-intrepid]

The benchmark result on my environment:

master:

test proto::h1::role::tests::bench_parse_incoming               ... bench:       3,199 ns/iter (+/- 1,142) = 274 MB/s
test proto::h1::role::tests::bench_parse_short                  ... bench:         282 ns/iter (+/- 10) = 141 MB/s
test proto::h1::role::tests::bench_server_encode_headers_preset ... bench:         246 ns/iter (+/- 13) = 439 MB/s
test proto::h1::role::tests::bench_server_encode_no_headers     ... bench:          34 ns/iter (+/- 0) = 2235 MB/s

this PR:

test proto::h1::role::tests::bench_parse_incoming               ... bench:       3,229 ns/iter (+/- 1,014) = 271 MB/s
test proto::h1::role::tests::bench_parse_short                  ... bench:         286 ns/iter (+/- 6) = 139 MB/s
test proto::h1::role::tests::bench_server_encode_headers_preset ... bench:         256 ns/iter (+/- 20) = 421 MB/s
test proto::h1::role::tests::bench_server_encode_no_headers     ... bench:          34 ns/iter (+/- 0) = 2235 MB/s

[seanmonstar]

Oh yea, so it does affect slightly bench_server_encode_headers_preset. Since this condition should be very rare, I think we can structure the change so that the normal case doesn't see any extra branches.

[ubnt-intrepid]

Fixed the patch to make the number of branch to be the same as the original.

benchmark result

master:

test proto::h1::role::tests::bench_parse_incoming               ... bench:       3,176 ns/iter (+/- 74) = 276 MB/s
test proto::h1::role::tests::bench_parse_short                  ... bench:         280 ns/iter (+/- 4) = 142 MB/s
test proto::h1::role::tests::bench_server_encode_headers_preset ... bench:         259 ns/iter (+/- 30) = 416 MB/s
test proto::h1::role::tests::bench_server_encode_no_headers     ... bench:          34 ns/iter (+/- 0) = 2235 MB/s

this PR:

test proto::h1::role::tests::bench_parse_incoming               ... bench:       3,193 ns/iter (+/- 107) = 274 MB/s
test proto::h1::role::tests::bench_parse_short                  ... bench:         280 ns/iter (+/- 8) = 142 MB/s
test proto::h1::role::tests::bench_server_encode_headers_preset ... bench:         245 ns/iter (+/- 20) = 440 MB/s
test proto::h1::role::tests::bench_server_encode_no_headers     ... bench:          34 ns/iter (+/- 0) = 2235 MB/s

[seanmonstar]

Thanks!