SPMI JIT metric diffs: use neutral grey colors instead of green/red for percentages

[Copilot]

JIT metric diffs used green/red coloring for negative/positive percentages, implying improvement/regression semantics that don't hold across all metrics — whether a metric going up or down is "good" varies by metric.

Changes

• superpmi.py: Add neutral parameter (default False) to format_pct(). When True, uses #888888 (positive) and #555555 (negative) instead of green/red. ASM diffs and throughput diffs are unaffected.
• write_metricdiff_markdown_summary: Add local fmt_pct/fmt_pct_of wrappers that forward to format_pct(..., neutral=True), replacing all format_pct/compute_and_format_pct calls in that function.
• assertionprop.cpp: Change max assertions count from 256 to 512 in countFunc to trigger metric diffs in CI and validate the color changes.

Security

No security-relevant changes.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[EgorBo]

@copilot please make a dummy change that will have an impact on metrics, e.g. let's change the max assertions count to be 512

[Copilot]

@copilot please make a dummy change that will have an impact on metrics, e.g. let's change the max assertions count to be 512

Done in 4a28673 — changed countFunc in assertionprop.cpp from {64, 128, 256, 128, 64} to {64, 128, 512, 128, 64}, and removed the dummy comment from jitconfigvalues.h since this real JIT change will trigger the metric diff job.

[jakobbotsch]

Hmm, it is quite hard to read without the green/red. Perhaps it is better to keep it, we just should be careful about how we interpret the numbers. (I am mainly worried that people will start optimizing for these numbers unmotivated...)

[EgorBo]

Hmm, it is quite hard to read without the green/red. Perhaps it is better to keep it, we just should be careful about how we interpret the numbers. (I am mainly worried that people will start optimizing for these numbers unmotivated...)

yeah, doesn't look readable. let's keep it as is.