Improved benchmark PR check & faster validation

[hmgaudecker]

Summary

Split the single asv-run-and-publish task into two purpose-specific workflows
and overhaul the PR benchmark comment format.

• PR branches: pixi run -e tests-cuda13 asv-run-and-pr-comment — runs benchmarks
  and posts a grouped comparison table as a PR comment
• Main branch: pixi run -e tests-cuda13 asv-run-and-publish-main — runs benchmarks
  and publishes to the dashboard

Benchmark naming & table format

• Rename benchmark classes to descriptive names (e.g. TimeMahlerYum → MahlerYum,
  TimeSolve → PrecautionarySavingsSolve) and standardize methods to generic names
  (time_execution, peakmem_execution, track_compilation_time)
• Rewrite pr_comment.py to parse ASV's pipe-separated markdown output into a grouped
  6-column table: Benchmark, Statistic, before, after, Ratio, Alert
• Alert column shows a cross mark when ratio > 1.10
• Commit hashes rendered as clickable links

Streamlined benchmark parameters

• Remove PrecautionarySavingsGridLookup (redundant with SimulateWithSolve)
• Add PrecautionarySavingsSimulateWithSolveIrreg for irregular grid coverage
• Bump subject counts: Simulate to 1M, SimulateWithSolve to 500k, Mortality to 100k
• Skip check_initial_conditions in benchmarks (measures simulation, not validation)

Vectorize validation (138x speedup)

Replace O(n) Python loops over JAX arrays in validation.py with vectorized
jnp.isin / jnp.unique / jnp.searchsorted operations. Validation for 100k
subjects drops from 30s to 0.2s. This makes large subject counts (1M) practical
for both benchmarks and user code.

Other fixes

• Fix hash collision: use head_sha_full[:8] instead of git rev-parse --short=8
• Paginate workflow comment listing (github.paginate)
• Handle missing gh CLI with helpful error message
• Backfill merge-base result file when benchmark names or params change
• Ensure HEAD result file exists under HEAD's hash when ASV ignores --set-commit-hash

CI workflow

The benchmark-check workflow checks for the <!-- benchmark-check --> comment
with embedded commit hash (current/stale/missing).

To satisfy the benchmark check

pixi run -e tests-cuda13 asv-run-and-pr-comment

Will be updated/automated once we have a server for this.

[read-the-docs-community]

Documentation build overview

📚 pylcm | 🛠️ Build #31919517 | 📁 Comparing c3dd9f4 against latest (ca71893)

🔍 Preview build

Show files changed (32 files in total): 📝 32 modified | ➕ 0 added | ➖ 0 deleted

File	Status
index.html	📝 modified
approximating-continuous-shocks/index.html	📝 modified
benchmarking/index.html	📝 modified
benchmarking-1/index.html	📝 modified
beta-delta/index.html	📝 modified
conventions/index.html	📝 modified
debugging/index.html	📝 modified
defining-models/index.html	📝 modified
dispatchers/index.html	📝 modified
function-representation/index.html	📝 modified
grids/index.html	📝 modified
index-1/index.html	📝 modified
index-2/index.html	📝 modified
index-3/index.html	📝 modified
index-4/index.html	📝 modified
installation/index.html	📝 modified
interpolation/index.html	📝 modified
mahler-yum-2024/index.html	📝 modified
mortality/index.html	📝 modified
pandas-interop/index.html	📝 modified
parameters/index.html	📝 modified
precautionary-savings/index.html	📝 modified
precautionary-savings-health/index.html	📝 modified
regimes/index.html	📝 modified
setup/index.html	📝 modified
shocks/index.html	📝 modified
solving-and-simulating/index.html	📝 modified
stochastic-transitions/index.html	📝 modified
tiny/index.html	📝 modified
tiny-example/index.html	📝 modified
transitions/index.html	📝 modified
write-economics/index.html	📝 modified

[hmgaudecker]

Benchmark comparison (main → HEAD)

Comparing ca718939 (main) → c3dd9f47 (HEAD)

Benchmark	Statistic	before	after	Ratio	Alert
Mahler-Yum	execution time	3.93±0.05s	3.71±0.01s	0.94
	compilation time	1.33m	1.32m	1.00
	peak mem usage	2.14G	2.14G	1.00
Mortality	execution time	145±4ms	144±1ms	0.99
	compilation time	7.47s	7.45s	1.00
	peak mem usage	1.2G	1.21G	1.00
Precautionary Savings - Solve (200)	execution time	19.1±0.6ms	20.1±3ms	1.05
	compilation time	2.74s	2.73s	0.99
	peak mem usage	1.03G	1.03G	1.00
Precautionary Savings - Solve (500)	execution time	19.4±1ms	21.3±1ms	1.09
	compilation time	3.33s	3.30s	0.99
	peak mem usage	1.04G	1.03G	1.00
Precautionary Savings - Simulate (1000000)	execution time	77.3±0.8ms	80.5±3ms	1.04
	compilation time	4.55s	4.56s	1.00
	peak mem usage	1.16G	1.16G	1.00
Precautionary Savings - Solve & Simulate	execution time	92.3±2ms	93.8±2ms	1.02
	compilation time	7.37s	7.40s	1.00
	peak mem usage	1.17G	1.17G	1.00
Precautionary Savings - Solve & Simulate (irreg)	execution time	422±1ms	422±1ms	1.00
	compilation time	8.23s	8.22s	1.00
	peak mem usage	1.22G	1.23G	1.00

[hmgaudecker]

BTW, we do now see a difference between grid types, as expected by @timmens:

Benchmark	Statistic	before	after	Ratio	Alert
Precautionary Savings - Solve & Simulate	execution time	92.3±2ms	93.8±2ms	1.02
	compilation time	7.37s	7.40s	1.00
	peak mem usage	1.17G	1.17G	1.00
Precautionary Savings - Solve & Simulate (irreg)	execution time	422±1ms	422±1ms	1.00
	compilation time	8.23s	8.22s	1.00
	peak mem usage	1.22G	1.23G	1.00

4.5-fold, but probably dwarfed in many models by other things.

@mj023, you may want to merge this branch into #280 for the timings already, so we'll have one harness for benchmarking going forward. Probably most useful to add a benchmark of the batched version to the precautionary savings model.