Expand SKMatrix tracking benchmarks for the managed-port surface (#4241)

[mattleibow]

What

Broaden SkiaSharp.Benchmarks.Tracking coverage of the SKMatrix surface that #4241 (port SKMatrix math to managed C#) improves. Today the only SKMatrix tracker is MatrixMapPointsBenchmark.MapPoints — the batch overload, which is the weakest indicator of #4241: the batch does a single native call for the whole array, so the per-call P/Invoke #4241 removes is amortised away. The single-op operations, where the managed↔native transition is paid every call, are where the win shows — and they had no permanent coverage.

Split the surface into two tracked groups, mirroring how #4241 changes it:

• MatrixMapPointsBenchmark (expanded — geometry mapping): add MapPoint, MapRect, MapVector, MapRadius, each looped over the existing deterministic point set. The pre-existing MapPoints method and its Points params are left untouched (a shipped history key must never be renamed). New methods are allocation-free; MapPoints keeps its allocating batch behaviour as a signal.
• MatrixOpsBenchmark (new — matrix algebra): Concat (pre + post multiply) and Invert — the ops that produce another matrix rather than mapping geometry.

Not added: MapVectors/MapPoints-batch duplicates (the existing batch MapPoints already represents that case) and the perspective path (#4241 falls back to native there — no gain to track).

Why these will move (and the batch barely will)

Local --job short vs the 4.151 nightly, per 4096-element workload — all zero-alloc:

Benchmark	Mean (4096)	Note
MatrixMapPointsBenchmark.MapPoints (batch)	1.88 µs	one native call — P/Invoke amortised, barely moves under #4241
MatrixMapPointsBenchmark.MapPoint	24.6 µs	~13× costlier/point than the batch = the P/Invoke #4241 removes
MatrixMapPointsBenchmark.MapRect	55.3 µs
MatrixMapPointsBenchmark.MapVector	57.8 µs
MatrixMapPointsBenchmark.MapRadius	118 µs	2 vectors/call
MatrixOpsBenchmark.Concat	125 µs	pre + post multiply
MatrixOpsBenchmark.Invert	71.5 µs

Since these paths are already zero-alloc, #4241's signal is a time drop; allocations stay flat at 0.

Constraints honoured

• Public API only (nightly NuGet ref); deterministic (fixed seed, no fonts/content).
• Affine operands (rotation/scale/translate) → Port SKMatrix math to managed C# to remove native interop (#2779) #4241's managed fast path.
• MatrixMapPointsBenchmark/MapPoints/Points unchanged (never-rename rule).
• Verified to compile against both the 3.119.4 baseline and 4.151 nightly — no new #if.

Follow-up: once this lands, update-branch #4241 so its pr column is measured against this expanded coverage.

[github-actions]

📦 Try the packages from this PR

Warning

Do not run these scripts without first reviewing the code in this PR.

Step 1 — Download the packages

bash / macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/mono/SkiaSharp/main/scripts/get-skiasharp-pr.sh | bash -s -- 4480

PowerShell / Windows:

iex "& { $(irm https://raw.githubusercontent.com/mono/SkiaSharp/main/scripts/get-skiasharp-pr.ps1) } 4480"

Step 2 — Add the local NuGet source

dotnet nuget add source ~/.skiasharp/hives/pr-4480/packages --name skiasharp-pr-4480

More options

Option	Description
--successful-only / -SuccessfulOnly	Only use successful builds
--force / -Force	Overwrite previously downloaded packages
--list / -List	List available artifacts without downloading
--build-id ID / -BuildId ID	Download from a specific build

Or download manually from Azure Pipelines — look for the nuget artifact on the build for this PR.

Remove the source when you're done:

dotnet nuget remove source skiasharp-pr-4480