Re-baseline the particle_render_10k budget to the CI reference toolchain (2.0 → 3.0 ms)

[offdev]

Root cause

Every merge to master since M2-PART-02 (PR #83) has failed the merge-lane CI Linux x64 (clang++) job on the particle_render budget gate (ParticleRenderBudget.TenThousandParticles) — and the unfiltered laige-render_tests entry that contains the same suite.

The gate platform is the CI reference lane (Clang 18.1.3, CMake Debug, ubuntu-24.04 shared runner), where the 10k-particle declare pass measures 1.91–2.68 ms mean (worst backend), consistently above the 2.0 ms bar:

CI run (master merge)	fpx16_16 mean	fp32_pinned mean
37780159655 (bdb323b)	2.67853	1.88209
37782008880 (2f2c5a9)	2.62795	1.94712
37850579155 (4f81866)	1.91031	2.07786

This is a calibration gap, not an engine regression: the 2.0 ms bar was set from local-machine runs (1.07798 ms canonical g++ 16.2.1 -O0, 1.18578 ms local Clang 23.1.1 -O0) and never verified with margin on the CI reference toolchain (runner Clang 18.1.3, not local 23.1.1, on a shared runner) — the exact lesson the M2-ISO-02 rebaseline recorded. The declareParticles path is unchanged since M2-PART-02, and the Linux x64 (g++) lane passed the same gated workload on the same commit.

The revision (methodology §5, the M2-ISO-02 rebaseline precedent)

• budgets.json: particle_render_10k target 2.0 → 3.0 (44% above the latest recorded reference-lane value 2.07786, 12.0% above the worst observed reference-lane mean 2.67853); measured 1.07798 → 2.07786 (latest recorded CI reference value, worse backend, run 37850579155).
• New baseline docs/benchmarks/baselines/m2-particle-render-budget-rebaseline.md (the eleventh baseline; supersedes m2-particle-render.md as the latest recorded value) + baselines index entry.
• Docs (DOC-003): docs/api/particle_render.md Budget bullet, src/laige-render/README.md, roadmap change log.
• No PRD revision — step-level budget (the depth_sort_10k precedent). No engine code, workload, or test change — budget + docs only (methodology §1: no regression occurred; this is a gate-margin calibration on the reference toolchain).

Verification

• Local (gated Linux non-instrumented trees, budget loaded at test time — no rebuild needed): ctest -R particle_render PASS on build, build-clang, build-release, build-shared; ctest -R budget_harness PASS (17-entry pin + well-formedness of the updated budgets.json).
• The definitive gate-platform check is the merge-lane Linux x64 (clang++) job (the PR lane runs Windows/macOS only — M0-CI-01); expected to PASS: the new 3.0 ms bar covers every observed reference-lane mean with ≥ 12% margin.