Fix Benchmark CI: green without running benchmarks since #489 (asv 0.6.5 vs py-rattler 0.22)

[glaziermag]

Every Benchmark / Linux run since Feb 4 has passed in ~20s without executing a single benchmark — 95 straight green runs (the first blind ones were #489's own checks). #489 bumped py-rattler to 0.22 in uv.lock, and asv 0.6.5's rattler plugin still calls solve(channels=...), which py-rattler >= 0.22 renamed to sources, so environment creation crashes immediately:

TypeError: solve() got an unexpected keyword argument 'channels'
·· Failure creating environment for rattler-py3.12

The failure check that should turn this red (benchmarks.yml lines 55-61 at d2857f3) never sees it: the traceback goes to stderr, bypassing the | sed | tee benchmarks.log pipe the grep reads, and tee masks asv's exit status (the step runs without pipefail). Example on main: run 30468384668 — success in 22s with the TypeError in its log. The last genuine runs are from Feb 4 morning, ~11-13 min each (e.g. 21679459393); every run from 21680736982 (Feb 4, 17:03 UTC) onward is blind.

Two commits:

1. benchmarks.yml — set -o pipefail + 2>&1, so an asv crash fails the job (and tracebacks get the existing ::error:: annotation).
2. uv.lock — uv lock -P asv==0.6.6: asv 0.6.6 handles the py-rattler rename, so benchmarks actually run again. (Dependabot's uv-group PRs bump py-rattler but not asv, so Bump the uv group across 1 directory with 8 updates #515 doesn't resolve this.)

Verified on my fork:

• A — unmodified main + no-op change: success, 25s, TypeError in log (current behavior).
• B — commit 1 only: the job now fails on the crash.
• C — both commits: benchmarks really run (28 min), job green, and the log ends with BENCHMARKS NOT SIGNIFICANTLY CHANGED.

I used Claude to find and bisect this; I re-ran the reproductions and read the logs and diffs myself.