WIP v3

[mgravell]

V3 tranche 1 is the reader-core, to resolve the OOM-related memory/corruption issue with V2

• merge the V3 core (Refactor IO "read" core logic #3014) on top of the V2/V3 pre-work (V3 pre-merge; all the early bits, for V2 #3028)
  • refactor all "read" logic to use RespReader instead of RawResult
  • nuke RawResult and CommandBytes
• consolidate refs (V2 SE.Redis does not have a RESPite ref for AsciiHash etc, so imports the files directly)
• resolve the stall on the 'sync'/'async' writer engines (the default 'pipe' reader engine is solid); this stall can generally be seen in the MovedTests which has a [Theory] usage for all-modes
  • emphasis for the reader: this stall is limited to modes that are only currently used in tests; they are the preferred option when stable, but I will not expose them until then
  • what this does: remove a thread-switch on the outbound path, by using the CycleBuffer core (think Pipe, but with discreet control over the read/write thread model); this includes the ability to use fully-sync dedicated writer; since tranche 1 is primarily reader-focused, I'm OK for this to slip unresolved to tranche 2, which is writer-focused
• remove trailing Pipelines.Sockets.Unofficial usage (marc/de-psu branch; compiles, needs testing)
• reimplement simulate-connection-failure
• replace the semaphoreslim replacement on netfx
• allow dedicated-thread sync read usage?
  • heh, turns out I forgot I already did this; it is tied to the write-engine (IsSync), so we know that it isn't part of the stall mentioned above (since that stalls on both 'sync' and 'async', and works fine for 'pipe')