fix(cron): add per-job memory_enabled override for memory provider access

[eulahwu917]

Problem

Cron sessions pass skip_memory=True to AIAgent, which prevents all cron jobs from using the memory tool or the external memory provider (e.g., hindsight). This is correct for most jobs (daily briefings, reports) but blocks cron jobs whose entire purpose is memory maintenance — like bepop-hindsight-memory-dump.

Root Cause

cron/scheduler.py hardcoded skip_memory=True at the AIAgent construction site (~line 3073). The memory provider plugin and MemoryStore are never initialized when skip_memory=True (agent/agent_init.py, lines 1335-1419), so the memory tool always returns \"Memory is not available.\" in cron sessions.

Fix

Add a per-job memory_enabled override following the same pattern as attach_to_session (already used in the codebase to opt cron jobs into session delivery):

File	Change
cron/scheduler.py	skip_memory=not bool(job.get(\"memory_enabled\")) — defaults True (backward compatible)
cron/jobs.py	Add memory_enabled: Optional[bool] = None to create_job(); persist only when explicitly set as bool
tools/cronjob_tools.py	Surface memory_enabled in both create and update paths of the cronjob tool
AGENTS.md	Document the opt-in override mechanism

Backward Compatibility

• Zero impact on existing jobs — the key is absent, so job.get("memory_enabled") returns None → skip_memory=True (existing behavior).
• Storage format unchanged — memory_enabled is only persisted when explicitly True or False, keeping existing jobs.json entries byte-identical.
• No merge or gateway restart required — this is a pure code change; the next cron tick picks it up.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #34098: both implement the same per-job memory_enabled opt-in through cron job creation/update and skip_memory in the scheduler. #34098 is the earlier open focused implementation.

[eulahwu917]

Reviewer — test coverage added per your revision request (12 new tests in `tests/cron/test_cron_memory_enabled.py`).\n\nWhile writing the tests, I also found and fixed 3 gaps in the original PR:\n- registry handler was not forwarding `memory_enabled` to `cronjob()` — tool schema calls would silently ignore it\n- CRONJOB_SCHEMA lacked the `memory_enabled` property\n- _format_job never surfaced `memory_enabled` when truthy\n\nAll 33 tests pass (21 existing workdir tests + 12 new memory_enabled tests, no regressions).\n\nReady for final review/merge.

[tonydwb]

Reviewed by Hermes Agent

[eulahwu917]

Closing per maintainer triage — confirmed duplicate of #34098, which is further along (passing CI, active review discussion, complementary PR #45769 already layered on top). Thanks for the fast triage.