fix(vm): release the store when Exec returns so a reused VM does not retain it - #211

Problem

vm.Exec binds the caller's store into the VM's RunState and only replaces it at the next execution. A VM that is kept and reused (the intended usage: the ledger caches one warm VM per script) therefore keeps its last run's store — and everything that store references — reachable after Exec returns.

In the ledger, that store reaches the apply scope of the Raft proposal that ran the script, including the whole preloaded coverage plan, so each cached VM pins an old proposal until the same script runs again or the entry is evicted.

Fix

Exec now defers runstate.Reset(nil): on every exit (success, execution error, or a panic raised by the store) the store is released along with the per-run balances. Results are unaffected — GetPostings already returns copies taken before the deferred reset.

Tests

• TestExecVmReleasesStore holds a weak pointer to the store and checks it becomes collectable while the VM is still alive, for the success, error and store-panic exits. It fails on all three without the fix.
• go test ./..., race on the touched packages, and golangci-lint are clean.

Performance

No measurable change (BenchmarkCompiledVM, BenchmarkCompiledVMCapped, BenchmarkCompiledVMFanIn, 5 runs each): identical allocations, timings within noise.