[FEAT] Microbenchmark add visualization

[Micky774]

Description

Introduces both static and interactive plotting visualization for microbenchmark results, including per-sample results.

Fixes # (issue)

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

New — benchmarks/microbenchmarks/

• viz_data.py — shared data layer: auto-detects the samples (--csv-samples) vs aggregate (--csv) schemas and normalizes both to one tidy frame; dtype shortening, label→bench/pass split, percentile trim, per-group stats.
• visualize.py — Plotly figure builders + CLI emitting self-contained interactive HTML (no server). Kinds: distribution (box + raw points), scaling (median vs M with spread band), bars (median throughput, IQR error bars), comparison (baseline-vs-candidate speedup bars), report (all). --cdn for a smaller file.
• dashboard.py — live Panel explorer (mirrors the JAX panel_app.py): selectors for plot/metric/x/hue/facet, percentile-trim slider, per-attribute filters; reuses visualize.py builders.
• requirements-viz.txt — pandas, plotly, panel.

Updated

• README.md — "Visualizing results" section (CLI + panel serve usage, supported benchmarks).

Notes

• Works for all microbenchmarks (dense/FP8/grouped GEMM, normalization, casting); pooled dimensions (e.g. grouped GEMM's B) trigger a [note] to split via --facet/--color/--pass.
• Speedup is direction-correct in both metrics (cand/base for throughput, base/cand for time; >1 = candidate faster).
• Read-only over the existing CSVs — no changes to the benchmark scripts or utils.py.

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[ipanfilo]

There are other PRs that add more scripts to microbenchmarks. So consider make this dir more structured: move visualization to subdir