Measure MILP overhead without trip count annotations

[byeongjee]

Adds a --no-tripcount flag to ckpt compile|bench|verify milp that skips the tripcount-annotation pass, so loops with bounds the pass cannot infer fall back to chunking.

Uses it to measure the cost of running without annotations at 10 µF (results/trip_count/):

• All 13 benchmarks compile; verify milp --no-tripcount passes 13/13.
• 8 benchmarks produce identical instrumentation, boundary hits, and execution time.
• The other 5 (bitcount, cuckoo_filter, dijkstra, rsa, stringsearch) slow down 2.15× geomean (5.11× max on dijkstra), with region boundary hits up 51× geomean.