Add failure persistence support for regression testing

[BowTiedRadone]

This PR implements failure persistence and regression testing for Rendezvous. When tests fail, the failing seeds and configuration used are now automatically saved to .rendezvous-regressions/, allowing users to replay them on subsequent runs to prevent regressions.

Key additions

• New --regr CLI flag to control test execution: if present, run only the regression tests
• Persistent failure storage grouped by test type (invariant/property) with automatic replay
• Failures sorted by timestamp for chronological tracking

Sample failure file
.rendezvous-regressions/ST1PQHQKV0RJXZFY1DGX8MNSNYVE3VGZJSRTPGZGM.counter.json:

{
  "invariant": [
    {
      "seed": 901717247,
      "dial": "example/sip010.cjs",
      "numRuns": 15,
      "timestamp": 1767886534833
    },
    {
      "seed": -1374686468,
      "numRuns": 9,
      "timestamp": 1767886531457
    },
    {
      "seed": 1298457354,
      "dial": "example/sip010.cjs",
      "numRuns": 20,
      "timestamp": 1767883389583
    },
  ],
  "test": [
    {
      "seed": 1656313995,
      "numRuns": 6,
      "timestamp": 1767886553125
    },
    {
      "seed": 64830639,
      "numRuns": 11,
      "timestamp": 1767886546477
    },
    {
      "seed": 1593583466,
      "numRuns": 3,
      "timestamp": 1767886542907
    },
  ]
}

This brings Rendezvous closer to production-grade fuzzing by ensuring discovered bugs stay caught, while improving internal code organization and maintainability.

Closes #130.

[BowTiedRadone]

Since --mode is now binary and --mode=new doesn't have any effect, a --regr option would be more suitable. To update.

[moodmosaic]

If a failure occurred at run 50, why force 100 runs? This could mask issues where failures only occur early in the sequence.

[BowTiedRadone]

This is an extra safety mechanism. In practice, it means a historical failure might have failed after 5 runs, but the same sequence of events (seed) can be considered passing if, for a given user configuration (seed, dial, etc.), it passes the default number of runs used by Rendezvous. However, you raised a great point; this behavior should probably be documented more clearly/explicitly.

[BowTiedRadone]

Let me know what you think, but please also consider the fact that only unique seeds are stored per test type (invariant/test). If a new failure happens for a different number of runs but same seed, it won't be persisted.