refactor(benchmarks): decouple CI baseline from analytics + unify measurements() (split prep)

[FBumann]

TODO (human): one line on why.

Note

Generated by AI (Claude).

Foundation for splitting the local analytics (plotting / sweep / memray engine / CLI) out of linopy into a standalone package — does the two structural changes that make that split clean, with no behaviour change.

Changes

1. Decouple the CI baseline from the analytics layer.
The CodSpeed baseline (registry + models/patterns + phases + test_*.py + conftest) now imports none of {memory, sweep, plotting, snapshot, bench, cli}. Two things were in the way:

• spec_param_id (the <name>-<axis>=<value> convention) moved from snapshot.py into registry.py — it's a core concept used by the drivers.
• benchmarks/__init__ no longer eagerly imports bench (from benchmarks import bench still works on demand).

Result: importing the package + the three CodSpeed test modules pulls in zero analytics; the dependency arrow only points analytics → core.

2. One measurements() contract for timing + memory.
The pytest drivers and the memray engine each re-derived "what runs + its node id" per phase — a duplication test_memory_id_alignment.py only band-aided. Now phases own phase_cases(phase) -> Iterable[PhaseCase] (each case a setup→action→teardown context manager); the drivers parametrize over it and the memray engine loops over it.

• Every benchmark node id is byte-identical (verified by an empty pytest --co diff) → CodSpeed baselines untouched.
• _measurements / _phase_tag gone; the memray engine now measures every available solver (was highs-only); the alignment test is deleted (ids are the same object now).
• pipeline becomes a first-class opt-in phase (--phase pipeline, both metrics; out of the default run + CI).

Verify

• ruff + mypy clean; harness tests pass; collected-node-id diff empty for every existing benchmark.

[codspeed-hq]

Merging this PR will not alter performance

✅ 79 untouched benchmarks
🆕 59 new benchmarks
⏩ 593 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	Memory	test_to_lp[kvl_cycles-severity=100]	N/A	38.5 MB	N/A
🆕	Memory	test_to_solver[highs-knapsack-n=10000]	N/A	2.5 MB	N/A
🆕	Memory	test_to_solver[gurobi-knapsack-n=10000]	N/A	2.8 MB	N/A
🆕	Memory	test_to_solver[gurobi-sos-n=1000]	N/A	3.1 MB	N/A
🆕	Memory	test_to_solver[highs-expression_arithmetic-n=250]	N/A	34.9 MB	N/A
🆕	Memory	test_to_solver[gurobi-piecewise-n=1000]	N/A	5.1 MB	N/A
🆕	Memory	test_to_solver[gurobi-qp-n=1000]	N/A	831.1 KB	N/A
🆕	Memory	test_to_solver[gurobi-kvl_cycles-severity=100]	N/A	198.8 MB	N/A
🆕	Memory	test_to_lp[nodal_balance-severity=100]	N/A	17.9 MB	N/A
🆕	Memory	test_to_lp[merge_balance-severity=100]	N/A	17.6 MB	N/A
🆕	Memory	test_build[milp-n=50]	N/A	283.7 KB	N/A
🆕	Memory	test_to_solver[highs-kvl_cycles-severity=100]	N/A	217 MB	N/A
🆕	Memory	test_build[masked-n=100]	N/A	735.4 KB	N/A
🆕	Memory	test_to_solver[highs-sparse_network-n=250]	N/A	64.8 MB	N/A
🆕	Memory	test_to_solver[highs-cumsum-severity=100]	N/A	70.7 MB	N/A
🆕	Memory	test_to_lp[sparse_network-n=250]	N/A	34.5 MB	N/A
🆕	Memory	test_to_solver[highs-masked-n=100]	N/A	659 KB	N/A
🆕	Memory	test_build[sos-n=1000]	N/A	444.1 KB	N/A
🆕	Memory	test_to_solver[highs-piecewise-n=1000]	N/A	4.2 MB	N/A
🆕	Memory	test_to_lp[milp-n=50]	N/A	63.3 KB	N/A
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---

_{Comparing refactor/benchmarks-decouple-core (d2a5d54) with master (85c103c)}

Footnotes

1. 593 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩