4.x: Improved parsing of HTTP/1 prologue and headers.

[tomas-langer]

Description

Improved parsing of HTTP prologue and headers.

• comparison through long words instead of bytes (faster CPU handling)
• the "fast path" for requests that do not need headers is improved, Connection header has optimized processing, as it is always used by Helidon webserver, and original code always had to use LazyString.stripOws which was sub-optimal; new version does a fast check for close and keep-alive connection options and bypasses this
• lazy headers are not copied when creating Headers, so the stripOws method is only called when the header is requested by user (or when validation is turned on)

[tomas-langer]

Relevant improvements (JMH tests run on my machine):

==================== Http1ParsingJmhTest.longHeader (+26.30%)
                              Baseline ████████ 5122185.288 ops/s
                               Current ███████████ 6950011.833 ops/s
==================== Http1ParsingJmhTest.multiBuffer (+20.60%)
                              Baseline █████████████████████ 13280652.339 ops/s
                               Current ██████████████████████████ 16726444.654 ops/s
==================== Http1ParsingJmhTest.singleBuffer (+27.63%)
                              Baseline ██████████████████████ 14260739.248 ops/s
                               Current ███████████████████████████████ 19704223.620 ops/s
==================== HttpPrologueParsingJmhTest.techEmpower (+8.27%)
                              Baseline ████████████████████████████████████████████████████████████████████████████████████████████ 58250706.875 ops/s
                               Current ████████████████████████████████████████████████████████████████████████████████████████████████████ 63503835.499 ops/s