Add convolution runtime regression benchmark

[Peter1169]

Summary

This adds a JMH runtime-regression benchmark for the convolution matrix described in the BoofCV bounty page:
https://boofcv.org/index.php?title=Bounties#Convolution_Runtime_Regression_Tests

Bounty recipient: Peter1169.

The benchmark covers:

• standard and normalized convolution paths
• horizontal, vertical, and 2D kernels
• kernel widths 3, 5, 7, and 21
• gray and interleaved image types, with 320x240 inputs and 3-band interleaved images
• reproducible randomized input data

It uses throughput mode so the regression output is reported as ops/sec.

Verification

Using local Azul Zulu JDK 25 toolchain provisioned for Gradle:

• ./gradlew autogenerate
• ./gradlew :main:boofcv-ip:compileBenchmarkJava
• ./gradlew spotlessCheck

I also ran short JMH smoke probes to confirm the benchmark is discovered and emits ops/sec:

• BenchmarkConvolveRuntimeRegression.horizontal, STANDARD, GRAY_U8, width 3: 6051.170 ops/s
• BenchmarkConvolveRuntimeRegression.vertical, NORMALIZED, INTERLEAVED_F64, width 7: 2412.284 ops/s
• BenchmarkConvolveRuntimeRegression.convolve2D, NORMALIZED, INTERLEAVED_U16, width 21: 16.279 ops/s