Blocked by grafana/flint#407

Summary

• replace Spotless Java formatting with Flint's native google-java-format check
• replace the custom StaticImportFormatter with Flint's generic regex-replace rule set
• replace Spotless license-header handling with Flint's native license-header check
• preserve the repository's long-string, formatter-off, static-import, and license-header behavior through Flint configuration
• pin Flint to the unreleased commit from the companion PR

Formatting baseline note

Some Java source changes in this PR are expected output changes from using google-java-format 1.35.0. They are formatter-version baseline changes, not additional Java logic changes; the main examples are the four benchmark-overhead test files.

Local timing

I measured this locally on a 24-vCPU environment using Temurin 25.0.3, with dependencies, tools, and Flint's native binary already installed. Timings are wall-clock medians of three runs unless noted otherwise. The ten-file rows use the same ten Java files in each tool.

For the ten-file rows, Spotless is ./gradlew spotlessJavaCheck and Flint is the complete mise run lint, so the Flint value includes all configured Flint checks, not only Java formatting. Flint has no persistent daemon; each value is a fresh native-process invocation. The formatter-only Flint checks took 0.052 s median when invoking the native binary directly.

The warm Spotless runs reused the configuration cache and had six Spotless tasks execute for the changed files; the cold-daemon runs also reused the configuration cache but started after ./gradlew --stop. This confirms that the daemon is a major part of the delay, but it is not the same thing as the configuration phase: the configuration cache controls reuse of configuration/task-graph work, while the daemon controls JVM/process reuse. The cold-checkout run also had to execute 1,242 tasks and store the configuration cache, so it is a worst-case first run rather than a daemon-only comparison.

The full Flint value is a genuine full-repository run using --full; it is not the changed-file baseline path. google-java-format accounts for 16.4 s of that run, while regex-replace took 0.873 s and license-header 0.037 s. The earlier 18.8 s value included mise exec overhead and is not used here.

The migration changes flint.toml and mise.toml, so this PR intentionally triggers a full Flint baseline against the pre-migration commit. Once that baseline is established, a typical ten-file edit is sub-second.

Companion PR

Depends on grafana/flint#407.

The mise.toml pin uses Flint commit 57a4c9b so this consumer PR can validate the unreleased native checks. It should be changed back to the released Flint backend after the Flint feature is released.

Validation

• mise run lint:fix (using the companion Flint commit)
• local Flint license-header, regex-replace, and google-java-format checks
• git diff --check