[BubbleCal]

The Linux x86-64 Java publisher repeatedly loses the final lance-jni compiler process to SIGKILL (#10726, #10773, #10788, and the beta.14 release run). Use the existing ubuntu-24.04-8x runner for this job, retaining the locked Debian 10 build, fat LTO, and all release platforms.

Validation on 409d1f6a5bbf5e8ad7daa25a2e99b7a871b0776e:

• Native dry run completed in 30m 44.85s with exit code 0 and peak RSS of 18,443,204 KiB (17.6 GiB). The final Rust 1.98.1 compiler command matches the failing beta.14 invocation byte for byte, including fat LTO and one codegen unit. No artifacts were uploaded or packages published by that run.
• Actionlint and git diff --check pass.
• Full PR CI passes Linux x86-64, Linux ARM64, macOS ARM64 native builds and Java 11 Maven packaging with all three native artifacts. Maven publication and R2 upload steps are skipped.

The memory measurement supports increasing runner resources; the original failures have no kernel OOM record, so their cause is not directly proven. This PR does not publish Maven/R2 packages.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

[lance-gatekeeper]

⚠️ Gate recommendation: approve with a non-blocking risk.

Using the larger runner established in #6175 is a reasonable, reversible mitigation for the reported JNI build failures while preserving the Debian 10 and fat-LTO release build.

The SIGKILL cause remains unconfirmed, so increased capacity may not eliminate the failures. If they recur, capture memory and termination diagnostics before further resource changes.