Tweak JIT params

[timfel]

Setting the trace_limit to sys.maxint gives me very nice traces through BitBlt >> #primDisplayString:from:to:map:xTable:kern: with 17263 ops. It calls into other loops plenty, but from the looks of it, having this loop removes a lot of bytecodes (as in - i can see pages of my terminal spanning 10 methods with just debug_merge_points and a single operation and guard near the end)

[krono]

I'm in for it. On pycket, we also use an increased trace limit due to the AST-interpreter nature. ( https://github.com/samth/pycket/blob/master/pycket/entry_point.py#L24 )

[timfel]

I noticed subjectively better interactive performance when I use all three adjustments from pycket.

trace_limit=1000000
threshold=131
trace_eagerness=50 

We need to look into this some more, maybe the master's project can do some benchmarking.

[timfel]

I can go up from 36,000,000 sends/sec (a fifth of Cog/Spur) to 320,000,000 sends/sec (almost twice Cog/Spur) on my machine by passing --jit max_unroll_recursion=16. Just goes to show how silly that benchmark is.