v2.6.1 — the code graph no longer occludes the brain

Bug fix. The 3D memory map was unusable in any repo that had a graphify code graph.

What was wrong

cml map auto-loaded graphify-out/graph.json from the working directory. Two problems fell out of that:

1. The map depended on where you ran it. Same command, same database, different picture.
2. In a graphified repo it hid the thing it exists to show. The code overlay occupies a single orbital slot — the same allocation as one project — while carrying up to 3000 nodes. Three thousand orbs packed into one slot render as a solid sphere sitting in front of everything else.

Measured in a real repo: 3398 nodes, of which 3000 were code. The 367 knowledge points that the durability gate exists to surface were a sliver squeezed off to the left, behind the ball. Context occluding the subject.

What changed

The code overlay is opt-in:

cml map                # memory only — 399 nodes, 1.6 MB
cml map --code         # opt in; auto-detects graphify-out/graph.json
cml map --code <path>  # point it anywhere

• --code with no graph present now says so rather than silently returning a different map.
• --no-code still parses, so existing invocations don't break.
• graphify already ships a dedicated viewer for code structure; this map is for memory.

Also

• The README no longer claims the map "picks up" graphify on its own, and its point count is refreshed to the measured 1,077 rows → 369 points.
• Statusline note (local, not shipped): if you drive a brain chip from cml stats, read the | N knowledge | field, not the leading N rows. Rows are the substrate — every indexed transcript row, most of it chatter. On this corpus that's the difference between showing 1077 and 369.

No schema change, no reindex needed.