ci: cap Java test shard heap

[ankitTelnyx]

Summary

• cap the production CI test-shard Gradle daemon at the previously working 2–8 GiB range
• keep the generated release configuration at 12 GiB for larger staging runners
• preserve all SDK, ProGuard, R8, and aggregate test coverage

Root cause evidence

After #219 removed the deterministic 30-minute timeout, release PR #216 still lost separate ubuntu-latest runners while using its newly generated -Xms12g -Xmx12g configuration:

• SDK shard: cancelled after 16m22s
• R8 shard: cancelled after 15m14s
• both steps ended with The operation was canceled, not test assertions

Current master uses -Xms2g -Xmx8g, and all three shards passed there. The release PR changes that file to 12/12 GiB, leaving too little host headroom on standard production runners.

Validation

• actionlint .github/workflows/ci.yml
• git diff --check
• all three shard task dry-runs passed with the command-line override
• exact release head 9a5b8fa8328df79a22ef442d4e2c401c909df690 precedence probe showed the daemon starting with -Xms2g -Xmx8g, overriding its generated 12/12 GiB properties