Replace blocking cl.launch by asynchronous cl.call.

[maleadt]

cl.launch, and its getindex-overloading shorthand, are currently blocking. That's pretty bad design when you want to launch as many kernels as possible to saturate the GPU. Besides, most subsequent API calls (like clEnqueueReadBuffer when doing a cl.read or copy!) are queue-ordered anyway, so I don't see why launches should be blocking.

Since this is a pretty subtle change, also rename launch to call to make sure nobody gets tripped up by this.

The goal is to build a high-level clcall and then @opencl on top of this, so the hope is that people won't have to deal with this.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 70.45%. Comparing base (2e243bc) to head (7b22dfb).
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #220   +/-   ##
=======================================
  Coverage   70.45%   70.45%           
=======================================
  Files           3        3           
  Lines         132      132           
=======================================
  Hits           93       93           
  Misses         39       39           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.