ENH: Add benchmark suite

[oyamad]

Summary

Add a benchmark suite for the package, so that performance-related PRs can
report standardized before/after comparisons.

• benchmark/benchmarks.jl: defines SUITE in the standard BenchmarkTools.jl
  format, compatible with PkgBenchmark.jl (and AirspeedVelocity.jl, if we later
  add CI benchmarking)
• benchmark/Project.toml: dedicated benchmark environment
• benchmark/README.md: usage instructions (standalone and via PkgBenchmark),
  modeled on GameTheory.jl's benchmark/README.md

What is benchmarked

Three model cases — 1-D stochastic optimal growth with Chebyshev and cubic
spline bases, and the 2-D Santos (1999) model from test/test_cdp_multidim.jl
with a quadratic spline basis — each with seven benchmarks:

Key	Description
s_wise_max_one_state	Per-state maximization kernel _s_wise_max!
bellman_operator	One application of bellman_operator!
compute_greedy	One application of compute_greedy!
evaluate_policy	One application of evaluate_policy!
set_eval_nodes	Evaluation on a non-interpolation grid
solve_PFI	End-to-end solve with PFI
solve_VFI_50iter	solve with VFI, capped at 50 iterations

Kernel benchmarks use basis coefficients from the converged PFI solution so
inputs are realistic; VFI is capped at 50 iterations so the benchmark measures
a fixed amount of work independently of convergence behavior.

Baseline (Julia 1.12.6, Apple M4 Pro)

Selected results from a verified PkgBenchmark.benchmarkpkg run, to indicate
current allocation behavior:

ID	time	memory	allocations
["1d_cheb", "s_wise_max_one_state"]	22.8 μs	135.6 KiB	1,102
["1d_cheb", "bellman_operator"]	778.7 μs	4.4 MiB	36,452
["1d_cheb", "solve_VFI_50iter"]	182.6 ms	700.3 MiB	5,826,322
["2d_spline", "evaluate_policy"]	1.2 ms	10.3 MiB	112,496
["2d_spline", "solve_VFI_50iter"]	182.3 ms	462.0 MiB	9,305,191

Usage

using PkgBenchmark
jud = judge("ContinuousDPs", "<target>", "<baseline>")
export_markdown("judgement.md", jud)

or standalone: julia --project=benchmark benchmark/benchmarks.jl
(see benchmark/README.md for details).

Verification

• Full suite runs cleanly via benchmarkpkg("ContinuousDPs") (~1.5 min)
• Standalone execution verified with --project=benchmark

🤖 Generated with Claude Code (Claude Fable 5)

[Copilot]

Pull request overview

This PR adds a dedicated benchmarking setup for ContinuousDPs.jl so performance-focused changes can be measured and compared consistently (standalone runs or via PkgBenchmark).

Changes:

• Added a BenchmarkTools.jl-style SUITE covering kernel-level operations and end-to-end solve runs for representative 1D/2D models.
• Introduced a dedicated benchmark environment (benchmark/Project.toml) to keep benchmark dependencies isolated.
• Documented how to run and compare benchmarks (standalone and via PkgBenchmark) in benchmark/README.md.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
benchmark/benchmarks.jl	Defines the benchmark SUITE with model cases and per-kernel / end-to-end benchmarks, plus standalone execution support.
benchmark/Project.toml	Adds a dedicated benchmark environment with the required dependencies.
benchmark/README.md	Provides usage instructions for running the suite and comparing commits with PkgBenchmark.