Optimize perf teardown performance

[GuillaumeLagrange]

Three things done here

1. Use bufwriters when writing to disk for debug_symbols
2. Serialize the whole perf-map content before saving in file for symbols
3. End fifo's loop faster after program termination

[codspeed-hq]

Merging this PR will not alter performance

✅ 4 untouched benchmarks

---

_{Comparing cod-1998-debug-symbol-collection-takes-a-very-long-time-in-walltime (5530881) with main (1ac1eae)}