fix(ruv-swarm): use fileURLToPath for module paths (fixes Windows C:\C:\ path bug)

[k2jac9]

Summary

Fixes broken module-path resolution on Windows in ruv-swarm caused by using new URL('.', import.meta.url).pathname to locate files relative to a module.

On Windows, new URL('.', import.meta.url).pathname returns a leading-slash, drive-prefixed string such as /C:/Users/.../ruv-swarm/src/. Feeding that into path.join(...) produces \C:\Users\..., and when an fs call later resolves that against the current drive you get the invalid doubled-drive path C:\C:\Users\....

Symptom

Every ruv-swarm invocation on Windows prints:

[ERROR] Failed to initialize pooled persistence {
  error: "ENOENT: no such file or directory, mkdir 'C:\\C:\\Users\\...\\ruv-swarm\\node_modules\\ruv-swarm\\data'"
}

The SQLite persistence layer never initializes (swarm state/memory is not saved), and the same broken construction is used for the WASM directory and the integrity-checksum path, so those are affected too.

Fix

Replace new URL(...).pathname with fileURLToPath(...) from node:url, which is the correct file: URL → filesystem path conversion. It produces a proper native path on Windows and is a no-op on POSIX (/home/... stays /home/...).

8 path constructions across 6 modules:

File	Constructions fixed
src/persistence.js	1 (default DB path)
src/persistence-pooled.js	1 (default DB path)
src/index.js	1 (WASM dir)
src/security.js	2 (checksums + dependency verify)
src/wasm-loader2.js	1 (base dir)
src/sqlite-pool.js	2 (worker script path)

Test evidence

Before: ruv-swarm init mesh 3 fails persistence init with the C:\C:\... ENOENT above.

After (Windows 11, Node 25):

🧠 Initializing ruv-swarm with WASM capabilities...
[INFO] Pooled persistence layer initialized successfully
✅ WASM bindings loaded successfully (actual WASM)
💾 High-availability pooled persistence layer initialized
📊 Pool configuration: 6 readers, 3 workers
🧠 Neural network capabilities loaded
📈 Forecasting capabilities loaded
✅ ruv-swarm initialized successfully
📦 Loading 0 existing swarms from database...
✅ Loaded 0 swarms into memory

No behavior change on POSIX. No dependency or API changes — purely a path-resolution correctness fix.

[k2jac9]

Quick context for triage: this hits every Windows user on ruv-swarm@1.0.20 — persistence fails to init on any invocation (e.g. ruv-swarm init mesh 3 → ENOENT ... mkdir 'C:\C:\...\data'), so swarm state/memory never persists, and the same .pathname pattern also breaks the WASM + checksum paths.

Root cause is just new URL('.', import.meta.url).pathname (yields /C:/... on Windows) → swapped to fileURLToPath(). No-op on POSIX, no API/dep changes — 8 spots across 6 files. Verified locally on Win11 + Node 25: persistence + WASM + neural all load clean after the change.

Happy to add a regression test or split per-file if you'd prefer.