perf: WAL synchronous=NORMAL — stop fsyncing every commit

lyra.db is on disk-backed ext4, where each WAL commit fsync'd (~0.15s here). Every
chat turn does several writes (remember user+assistant, summaries, poker logging),
so this was adding real per-turn latency, and made the dream loop + tests crawl.
synchronous=NORMAL is WAL's recommended companion: durable across app crashes, only
a power/OS crash can drop the last txn (never corrupts). Per-write dropped from
~0.4s to ~0.001s; test suite 72s -> 24s.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

lyra/memory.py

@@ -131,6 +131,11 @@ def _connection() -> sqlite3.Connection:
         # alongside the web server without tripping "database is locked".
         _conn.execute("PRAGMA busy_timeout=5000")
         _conn.execute("PRAGMA journal_mode=WAL")
+        # WAL's recommended companion: don't fsync on every commit (only at
+        # checkpoint). Safe against app crashes; a power/OS crash can lose the last
+        # txn but never corrupt. On disk-backed storage this turns ~0.15s-per-commit
+        # fsync latency into ~nothing — big win for per-turn writes + the dream loop.
+        _conn.execute("PRAGMA synchronous=NORMAL")
         _conn.executescript(SCHEMA)
         # Migrations for DBs created before a column existed (no-op if present).
         for ddl in ("ALTER TABLE sessions ADD COLUMN mode TEXT",):