Bug: apc memory add uses legacy entry_id schema while apc collect uses new id/source_file schema — pollutes cache with mixed schemas

[FZ2000]

Summary

apc memory add creates entries using the old entry_id schema, while apc collect creates entries using the new content-hash id/source_file schema. This causes a mixed-schema cache where apc status shows inflated and confusing memory counts.

Steps to Reproduce

# Add some manual memory entries
apc memory add "prefer Python type hints"
apc memory add "use pytest for testing"

# Run collect
apc collect -y

# Check status
apc status
# Shows: Memory: 197 (or whatever large number)
# But you only added 2 entries manually + collect found 6 files!

Actual Behavior

• apc memory add stores entries with schema: {"entry_id": ..., "category": ..., "content": ..., "source": "manual_add"}
• apc collect stores entries with schema: {"id": ..., "source_tool": ..., "source_file": ..., "label": ..., "content": ..., "collected_at": ...}
• Both schemas coexist in ~/.apc/cache/memory.json
• apc status reports the total count of ALL entries (both schemas combined) as a single "Memory" number — very confusing

In this test environment with 195 manually-added test entries (from Docker integration tests), apc status shows Memory: 203 even though there are only 8 actually-collected memory files. New users running apc memory add followed by apc collect will see memory counts far higher than expected.

Expected Behavior

Option A: Migrate apc memory add to use the new id-based schema for consistency.

Option B: apc status should separately report:

• Collected files: 8 (from apc collect)
• Manual entries: 195 (from apc memory add)

Option C: apc memory add should clearly be documented as a legacy command that creates a different type of entry.

Affected Code

# src/memory.py — add() command still uses old schema:
new_entry = {
    "entry_id": entry_id,    # OLD schema
    "category": category,
    "content": text,
    "source": "manual_add",
}

Environment

• macOS Darwin 25.2.0 (arm64)
• apc version 0.1.0

Severity

Medium — Creates user confusion about memory count and makes it hard to audit what's actually being synced to AI tools.