Knowledge stores raise on Z-suffixed timestamps (Python 3.10); consolidate ISO-parse into a shared _time.py

[johnnichev]

Bug

sessions.py (line ~1071) strips a trailing Z from ISO timestamps before parsing:

dt = datetime.fromisoformat(str(value).replace("Z", "+00:00"))

…but the knowledge stores call datetime.fromisoformat(...) raw, without the Z handling:

• src/selectools/knowledge.py:294,297,470,473,567,575
• src/selectools/knowledge_store_supabase.py:72,75
• src/selectools/knowledge_store_redis.py:72,75

On Python 3.10, datetime.fromisoformat("2026-01-01T00:00:00Z") raises ValueError (the Z suffix is only accepted on 3.11+). Supabase/Postgres timestamptz columns can serialize with a Z, so reading such a row back raises on 3.10.

Fix (also a cleanup)

The fromisoformat + naive→UTC-coercion idiom is hand-rolled ~18× across knowledge.py, the knowledge stores, unified_memory.py, and sessions.py. Consolidate into a small src/selectools/_time.py:

def ensure_aware(dt):  # naive -> UTC
    return dt if dt.tzinfo is not None else dt.replace(tzinfo=timezone.utc)

def parse_iso_utc(value):
    return ensure_aware(datetime.fromisoformat(value.replace("Z", "+00:00")))

Replace the hand-rolled sites with parse_iso_utc / ensure_aware. This fixes the Z bug everywhere in one place.

Acceptance

• A knowledge store round-trips a …Z-suffixed timestamp without raising on Python 3.10.
• The ~18 hand-rolled parse sites call the shared helper.
• A regression test covers the Z-suffix parse.

[johnnichev]

Fixed in #151 (merged to main). All 17 ISO-parse sites now route through selectools._time.parse_iso, which normalizes the trailing Z before fromisoformat. Regression coverage added in tests/test_time.py.