fuzz, txncache: add transaction cache fuzzer

[two-heart]

Keeps a model to track invariants. Finds #9517

[greptile-jt]

Prompt To Fix All With AI

This is a comment left during a code review.
Path: src/flamenco/runtime/fuzz_txncache_fork_graph.c
Line: 129

Comment:
**Missing `seed` argument to `fd_txncache_shmem_new`**

`fd_txncache_shmem_new` takes 4 parameters (`shmem`, `max_live_slots`, `max_txn_per_slot`, `seed`), but this call only passes 3. Every other call site in the codebase (e.g. `test_txncache.c`, `fd_svm_mini.c`) passes `0UL` as the seed. This will cause a compilation error.

```suggestion
  fd_txncache_shmem_t * shtc = fd_txncache_shmem_join( fd_txncache_shmem_new( fuzz_shmem, max_live_slots, max_txn_per_slot, 0UL ) );
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/flamenco/runtime/fuzz_txncache_fork_graph.c
Line: 836-838

Comment:
**Wrong fork_id passed to `fd_txncache_insert`**

`fd_txncache_insert`'s first fork argument is the *executing* fork (the fork the transaction ran on), not the ancestor holding the blockhash. Here `block_fork_id` (the ancestor) is passed instead of `fork_id` (the executing fork).

This causes two problems:
1. If `block_fork` is `FORK_FINAL` (frozen==2), the assertion `FD_TEST( fork->shmem->frozen<=1 )` inside `fd_txncache_insert` will fire.
2. If `block_fork` is `FORK_HASHED` (frozen==1), the transaction is stored with the wrong `fork_id`, causing model/target query mismatches since the model records `txn_fork = fork_id` while the target stores `fork_id = block_fork`.

Compare with `model_insert_one` (line 859) which correctly passes `fork_id`:

```suggestion
  fd_txncache_insert( m->tc, (fd_txncache_fork_id_t){ .val = fork_id }, m->fork[ block_fork ].blockhash.uc, txnhash->uc );
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fuzz, txncache: add transaction cache fu..." | Re-trigger Greptile}

[greptile-jt]

_{Reviews (2): Last reviewed commit: "fuzz, txncache: add transaction cache fu..." | Re-trigger Greptile}

[intrigus-lgtm]

I assume this is supposed to model a probability distribution?

[intrigus-lgtm]

Any reason to not move check_invariants to the txncache directly as e.g. done for fd_rdisp or fd_forest?