performance: use first h2 frame header detect instead of read first frame from whole payload

[yuweizzz]

fix #700

the h2 response detect will read whole payload, which may cost too much memory.

[cfc4n]

@chilli13 Can you test the performance?

[chilli13]

@chilli13 Can you test the performance?

The optimization effect on memory consumption is quite obvious.
50MBps http2 traffic
taskset -c 8 ./bin/ecapture tls

before optimism (master)

=====================
Process ecapture resource consumption statistics
CPU Average: 128.20%
CPU Max: 152.00%
Memory Average: 629.41 MB
Memory Max: 694.56 MB
=====================

after optimism (h2-patch)

=====================
Process ecapture resource consumption statistics
CPU Average: 72.60%
CPU Max: 81.00%
Memory Average: 389.11 MB
Memory Max: 508.99 MB
=====================

[yuweizzz]

Why the data[4] "works":

data[4] actually read the frame flags, look back for flags type:

const (
	// Data Frame
	FlagDataEndStream Flags = 0x1
	FlagDataPadded    Flags = 0x8

	// Headers Frame
	FlagHeadersEndStream  Flags = 0x1
	FlagHeadersEndHeaders Flags = 0x4
	FlagHeadersPadded     Flags = 0x8
	FlagHeadersPriority   Flags = 0x20

	// Settings Frame
	FlagSettingsAck Flags = 0x1

	// Ping Frame
	FlagPingAck Flags = 0x1

	// Continuation Frame
	FlagContinuationEndHeaders Flags = 0x4

	FlagPushPromiseEndHeaders Flags = 0x4
	FlagPushPromisePadded     Flags = 0x8
)

those value uint32(data[4])>>7 will always pass the check.

[cfc4n]

LGTM, Thanks.