perf: batch HPACK encoder writes per u32

[seanmonstar]

This accumulates HPACK encoding into a u32, and then writes to the buffer with put_u32, instead of put_u8 for each byte. Seems to improve HPACK encoding performance a nice amount:

Input	Before	After	Improvement
www.example.com	94.67 ns	58.07 ns	39% faster
Typical header value	401.02 ns	169.01 ns	58% faster
Long ASCII value	564.40 ns	272.63 ns	52% faster
All 256 octets	4.603 µs	1.649 µs	64% faster