bench: tensorstore CPU vs damacy GPU read+decode comparison

[nclack]

Adds bench/tensorstore_bench.py, a scenario-driven CPU read+decode benchmark
using tensorstore, so the GPU (damacy) vs CPU (tensorstore) tradeoff for zarr v3
read+decode can be measured on identical data, chunking, and patch sampling.

• Consumes the existing Scenario JSON schema via bench/scenario.py (same as
  run.py), handling both the synthetic path (uris is None, arrays enumerated
  from uri_fmt/array_path/n_zarrs) and explicit uris.
• Opens arrays with tensorstore's zarr3 driver; skips wrong-rank/too-small arrays
  with the same filtering as the bench so array counts line up.
• Ports bench/main.c's xorshift64* RNG and draw order (array index, then
  per-axis start), so with a shared seed the sampled patches match damacy's
  bit-for-bit and both read the same bytes.
• --threads concurrency sweep (default 1,2,4,8,16,32) via tensorstore context
  limits + a bounded in-flight read window, reporting samples/s and GB/s per
  thread count and the best point — the CPU thread pool is the real comparison
  point against a single GPU decode stream.
• --drop-cache mirrors run.py's page-cache drop for cold-read measurement;
  --compare-with <damacy results.json> prints a head-to-head line.
• Emits a table plus a JSON summary shaped like the existing bench output.

tensorstore fuses read+decode, so only total throughput is reported (no per-stage
split). Smoke-tested on a synthetic scenario and a uris scenario on a login node;
the full cold sweep runs on a compute node.

Closes #153.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 57.73%. Comparing base (17239ae) to head (537d29b).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #155   +/-   ##
=======================================
  Coverage   57.72%   57.73%           
=======================================
  Files          64       64           
  Lines       10055    10055           
  Branches     1750     1750           
=======================================
+ Hits         5804     5805    +1     
+ Misses       3501     3500    -1     
  Partials      750      750           

Flag	Coverage Δ
unittests	57.73% <ø> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.