testing/ostest: shrink default round-robin getprime test duration

[Swatantra-66]

Summary

The round-robin scheduler test in testing/ostest/roundrobin.c executes two concurrent threads calculating prime numbers up to CONFIG_TESTING_OSTEST_RR_RANGE across CONFIG_TESTING_OSTEST_RR_RUNS.

With the previous defaults of range 30,000 and 10 runs, each thread performed over 2.25 billion trial-division loop iterations. On resource-constrained microcontrollers (such as Cortex-M0/M3, Raspberry Pi Pico / RP2040, or AVR), this causes OSTest to stall on the getprime step for 5+ minutes.

This PR:

1. Restores default CONFIG_TESTING_OSTEST_RR_RANGE from 30000 to 10000 (re-aligning with the file header note in roundrobin.c stating default values target ~30s on Cortex-M3 @ 24MHz, and matching the official documentation in Documentation/applications/testing/ostest/index.rst).
2. Reduces default CONFIG_TESTING_OSTEST_RR_RUNS from 10 to 4 (and updates the fallback #define in roundrobin.c accordingly).

4 runs across 2 threads still produce 8 recorded context-switch samples in g_rr_values, providing ample opportunity to detect round-robin interleaving while cutting execution time by over 90%.

Fixes #3620

Impact

• Significantly accelerates ostest execution on embedded targets and automated test runs.
• Prevents 5-minute stalls during release qualification testing on boards like Raspberry Pi Pico.
• No breaking change: targets requiring longer stress tests can still customize both options via Kconfig/defconfig.

Testing

• Verified Kconfig syntax and options range validation [1; 32767].
• Verified roundrobin.c fallback preprocessor definitions and compilation consistency.

[Swatantra-66]

Looks like Linux (risc-v-04) hit a transient ghcr.io Docker login timeout (context deadline exceeded). Could a maintainer please re-trigger the failed check?
Thank You!

[Swatantra-66]

The single failure in Linux (arm64-01) appears to be an intermittent QEMU SMP crash during test_os (test_getprime passed 100%).

[raiden00pl]

@xiaoxiang781216 @Swatantra-66 this PR broke CI: https://github.com/apache/nuttx-apps/actions/runs/37354726799/job/112474565627

[Swatantra-66]

Thanks for investigating and catching this, @raiden00pl!

When arm64-01 failed during the PR checks, I had assumed it was an intermittent QEMU crash since standalone test_getprime passed. I didn't realize that ostest's internal rr_test was finishing all 4 runs within a single time slice on fast virtual CPUs like qemu-armv8a, tripping the round-robin assertion.

Thank you @casaroli for submitting the revert in #3832 to restore CI. Going forward, we should adjust the range/runs only in individual board defconfigs (like RP2040) rather than changing the global Kconfig defaults.

[linguini1]

These kinds of changes should have runtime tests.