gh-146073: Add fitness/exit quality mechanism for JIT trace frontend

[cocolato]

Reopen: #147966

• Issue: Improving trace quality by tracking "fitness" and "exit quality" #146073

[cocolato]

It appears that the current parameters do not yet guarantee runtime safety; I will continue to work on fixes and optimizations.

[markshannon]

I've commented on the issue #146073 (comment)

[cocolato]

The new branch's bench result and executor graph (Debian 13 WSL2, Linux 6.6.87.2, x86_64， AMD Ryzen 5 5600X):

+----------------------+----------+-----------------------+
| Benchmark            | baseline | fitness               |
+======================+==========+=======================+
| raytrace             | 251 ms   | 222 ms: 1.13x faster  |
+----------------------+----------+-----------------------+
| pickle_pure_python   | 275 us   | 255 us: 1.08x faster  |
+----------------------+----------+-----------------------+
| go                   | 84.1 ms  | 79.2 ms: 1.06x faster |
+----------------------+----------+-----------------------+
| spectral_norm        | 54.5 ms  | 52.2 ms: 1.05x faster |
+----------------------+----------+-----------------------+
| regex_compile        | 101 ms   | 98.7 ms: 1.03x faster |
+----------------------+----------+-----------------------+
| unpickle_pure_python | 173 us   | 171 us: 1.01x faster  |
+----------------------+----------+-----------------------+
| pidigits             | 137 ms   | 136 ms: 1.01x faster  |
+----------------------+----------+-----------------------+
| regex_effbot         | 1.95 ms  | 1.97 ms: 1.01x slower |
+----------------------+----------+-----------------------+
| xml_etree_process    | 49.0 ms  | 49.5 ms: 1.01x slower |
+----------------------+----------+-----------------------+
| xml_etree_generate   | 74.5 ms  | 75.4 ms: 1.01x slower |
+----------------------+----------+-----------------------+
| json_dumps           | 7.31 ms  | 7.42 ms: 1.01x slower |
+----------------------+----------+-----------------------+
| nbody                | 50.4 ms  | 51.2 ms: 1.02x slower |
+----------------------+----------+-----------------------+
| fannkuch             | 258 ms   | 262 ms: 1.02x slower  |
+----------------------+----------+-----------------------+
| json_loads           | 18.4 us  | 18.8 us: 1.02x slower |
+----------------------+----------+-----------------------+
| xml_etree_parse      | 118 ms   | 120 ms: 1.02x slower  |
+----------------------+----------+-----------------------+
| deltablue            | 2.19 ms  | 2.26 ms: 1.03x slower |
+----------------------+----------+-----------------------+
| pyflate              | 272 ms   | 282 ms: 1.03x slower  |
+----------------------+----------+-----------------------+
| richards             | 16.1 ms  | 16.9 ms: 1.05x slower |
+----------------------+----------+-----------------------+
| Geometric mean       | (ref)    | 1.00x faster          |
+----------------------+----------+-----------------------+

For the richards test, reducing the branch penalty can help us close loops.