feat: wal durability

[notxorand]

closes: #21

Greptile Summary

The PR adds configurable eventual or strict WAL durability and integrates durable source and cascade checkpoints into runtime execution.

• Adds WAL-backed fact storage, recovery, batching, and durability modes.
• Routes inference cascades through durable pending work and atomic cascade checkpoints.
• Adds rollback handling for failed cascade mutations and queue scheduling.
• Integrates shared storage into CLI and supervisor-managed sessions.
• Adds storage and WAL tests and benchmarks.

Confidence Score: 3/5

The PR is not safe to merge until cascade rollback preserves unrelated queued work and strict WAL publication resolves terminal failures without hanging.

Cascade rollback currently deletes all shared dirty signals, while a persistent post-fsync publication failure traps the WAL worker and strict caller in an unbounded retry loop.

Files Needing Attention: src/engine/context.zig and src/wal.zig

Important Files Changed

Filename	Overview
src/wal.zig	Implements WAL admission, recovery, cascade checkpoints, and strict publication, but strict publication can retry forever after a synced commit.
src/engine/context.zig	Adds cascade accumulation and rollback, but rollback clears unrelated work from the shared dirty queue.
src/engine/loop.zig	Drives durable source work and transitive dirty entries through bounded inference cycles with convergence checks.
src/wasm/host_abi/generic.zig	Integrates guest writes with cascade accumulation, queue signaling, and rollback ownership.
src/storage.zig	Exposes WAL-backed persistence and configurable durability to engine callers.
bin/supervisor.zig	Opens shared durable storage and supplies it to managed module sessions.

_{Reviews (20): Last reviewed commit: "fix: keep cascade and WAL state recovera..." | Re-trigger Greptile}

Greptile also left 2 inline comments on this PR.

[codspeed]

Merging this PR will degrade performance by 11.97%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 1 regressed benchmark
✅ 3 untouched benchmarks
🆕 1 new benchmark

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	runtime.execution	7.4 s	8.4 s	-11.97%
🆕	runtime.storage	N/A	2.1 ms	N/A

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing wal-durability (52ab58f) with dev (89658db)}