Remove SlabAllocator

[trevnorris]

Removing the SlabAllocator will allow for more aggressive performance enhancements.

EDIT: I have removed a couple commits from the set that weren't directly related. While this does introduce a small performance regression for requests of small packet size, it seems make bench is exasperating that discrepancy. If the benchmarks are run by hand (running benchmark/http_simple.js and pointing wrk to the appropriate url) the performance difference is easily below 5%. And this is only for small (i.e. <= 1024 bytes) requests. This patch is a launchpad to several other changes. Some of which include:

• Allowing the character encoding to propagate down to StreamWrap to remove any redundant object creation.
• Give more flexibility to StreamWrapCallbacks::DoRead callbacks so objects (either buffers or external string resources) don't need to duplicate memory before being used.
• Use those custom callbacks to internally buffer key items, such as HTTP requests. This is advantageous because we'd drop the step of creating a buffer then passing it back to http_parser. Instead we accumulate until all headers have been received.
• Easily create data "pass along" (I'm sure there's a more correct term for this). Like a pipe, except after pointing the source to the destination in your script the data itself never needs to reach the js level.

[trevnorris]

sometimes I hate you github!

[trevnorris]

The pr summary has been updated to reflect current changes. imo I'd like just these changes merged first. Mainly because there's no finish line of performance improvements it offers (as outlined in the summary above) and I don't want to go through another massive buffer-type pr scenario again. :)

/cc @isaacs @bnoordhuis

[bnoordhuis]

Prefer braces around statement when it's spread over multiple lines.