t27c lab: /data fills with idle session target dirs and stops the queue #7357

Problem

The t27c lab's /data volume (46G) filled to 99% on 2026-10-07. The worker thread died on ENOSPC while recording a failed run and the enqueue queue stood still from about 06:41Z (worker fix landed in #7351, which closed #7090).

What filled it

• /data/ref/*/w0..w11 and wt: worker scratch and worktrees of finished reference experiments (each has a DONE marker). About 0.9G per experiment, 11 experiments.
• /data/<name>-target, /data/ref/target*, /data/scratch-t27a/*target: cargo target dirs left behind by sessions, 130-725M each, most untouched for 1-3 days.
• /data/ref/zctx-*: zig cache temp dirs.

Done by hand

Removed only the dirs above that no live process used and that had no file newer than 12 hours. 17.7 GiB freed: 45G used, 694M free (99%) -> 28G used, 18G free (61%). /data/src, /data/target, /data/www, /data/cargo, /data/zig and the three fresh target dirs were left alone.

Fix

The lab's poller drops /data/*-target dirs untouched for a day whenever free space falls under 15%. The lab's own /data/target is not matched. Covered by lab.py --self-check.