Optimize server2

[malloch]

Server code reorganization & consolidation, fixes some erroneous behaviour and hopefully makes future maintenance easier:

• use lo_servers_wait internally for all calls to poll/select, i.e. with a 0 timeout for lo_server_recv. This removes a bunch of duplicate code
• don't call poll/select again if we already know something was received
• return an error (-1) from lo_server_wait and lo_servers_wait if returning early due to error rather than msg received or full timeout
• do not return from lo_server_recv if a received message was queued for future dispatch
• if a queued message is ready for dispatch, dispatch it rather than just queuing a new message and returning

Adds a configure/cmake option for disabling poll if desired since testing shows a substantial performance penalty for using this code path in liblo (possibly due to the alloca?). On MacOS/M2 Pro, the testspeed test in libmapper completes in ~4.3 seconds using the liblo poll path and ~2.6 seconds using select.

[radarsat1]

No objection here, if you want to take it out of draft I will merge it. (Although I find it very surprising that either poll or alloca would cause any latency.)

[malloch]

Apparently the slow poll is a BSD/MacOS thing, with the only solution to use kqueues instead: https://forums.developer.apple.com/forums/thread/764698

[malloch]

I have another upcoming commit to add to the PR then I will take it out of draft.