fix(crate_universe): pass OUTPUT_BASE as --output_base startup flag to bazel info

[georgesfarah]

Problem

The OUTPUT_BASE env var was added to let CI environments supply the output_base without relying on bazel info. However, the override is applied after parsing the bazel info output:

let output = process::Command::new(bazel).arg("info")...output()?;
if !output.status.success() {
    bail!(output.status)    // <-- bails here when output_base is not writable
}
// ... parse output ...
if let Ok(path) = env::var("OUTPUT_BASE") {  // <-- never reached
    bazel_info.insert(...);
}

When bazel info fails (e.g. the default ~/.cache/bazel/... is not writable in a sandboxed CI), the function bails before OUTPUT_BASE is ever checked — making the env var useless for its intended purpose.

Fix

Pass OUTPUT_BASE as a --output_base Bazel startup flag (before the subcommand):

bazel --output_base=/path/to/base info release output_base

Startup flags take effect before Bazel accesses any path, so it never touches the default (possibly non-writable) output_base. The bazel info subprocess runs normally and the post-parsing override is no longer needed.

Changes

• crate_universe/src/cli/vendor.rs: restructure BazelInfo::try_new to optionally inject --output_base as a startup flag; remove the post-parsing OUTPUT_BASE override.
• Removed test_parse_bazel_info_output_base_env_override since parse_bazel_info no longer handles the OUTPUT_BASE override — the existing test_bazel_info already covers the parsing logic.

[UebelAndre]

Good catch, thank you!