Measure whether mise install serialization is still needed

[junior-ricon]

Problem

Template has set MISE_JOBS=1 since the initial tool skeleton in #1, but the introducing PR and commit do not document a concrete concurrency failure. The setting serializes mise's cold tool installation and may add avoidable setup time.

Template#5 narrows the variable from the whole job to the mise-action install step. That is safer than leaking install policy into tests and lints, but it deliberately preserves the unexplained serialization until it can be measured.

Investigation

• Establish repeated cold-install baselines on Ubuntu and macOS with MISE_JOBS=1.
• Remove the override and repeat the same cold matrix enough times to expose intermittent plugin/tool installation races.
• Compare setup duration and inspect failures or shared-state corruption around vfox-shiv and the floating shiv:* tools.
• Keep cache-hit behavior separate from cold-install behavior; warm runs perform little installation and do not answer whether serialization is needed.

Completion

• If concurrent installs remain reliable and improve or preserve setup time, remove MISE_JOBS=1 from Template.
• If concurrency fails, preserve the scoped override and document the exact reproduced failure it prevents.
• Record commands, run IDs, platform results, and before/after timing.

This is separate from #4, which concerns parallel execution of BATS after tool installation.