Add a repeatable generated-shim stress matrix

[junior-ricon]

Problem

Generated shims are executable artifacts assembled from install-time paths, package names, embedded shell, and runtime environment. Ordinary unit and integration coverage catches known regressions, but it does not systematically pressure the boundary between generation-time state and runtime state.

The investigation and self-review for #146 / #148 used useful ad hoc stresses:

• cache paths containing spaces, apostrophes, and literal shell metacharacters;
• populated and missing install-local task maps with a conflicting runtime/global map;
• two same-named package versions with separate generated shims and maps;
• bash -n over the generated artifact;
• generation under one XDG cache and execution under another.

These are repeatable generated-artifact contracts rather than one-off debugging tricks.

Proposed shape

Add a deterministic generated-shim stress matrix, preferably using the existing mise run test [suite] interface rather than creating a parallel test runner.

A likely shape is test/shim-stress.bats, included in the default suite if its runtime stays modest and targetable with:

mise run test shim-stress

Cover a bounded matrix such as:

1. Hostile but valid paths
   • spaces;
   • apostrophes;
   • dollar signs and backticks treated literally;
   • nested install/cache roots.
2. Cache lifecycle
   • populated map;
   • missing map regenerated on first execution;
   • conflicting runtime XDG/global map;
   • same package name under two install roots.
3. Artifact validity
   • bash -n on each generated shim;
   • execute a real nested mise task through the generated shim;
   • assert writes stay inside the owning install/cache root.
4. Environment drift
   • generation and execution with different XDG_CACHE_HOME values;
   • inherited mise overrides do not contaminate package task discovery.

Run it on Ubuntu and macOS. If the matrix becomes too slow for every default test run, give it a dedicated CI job on lib/shim.sh, cache, resolver, install/update, and stress-test changes, plus the existing scheduled workflow. Keep cases deterministic; if seeded fuzzing is added later, print and preserve the seed for reproduction.

Boundaries

• Do not duplicate every normal shim test in a second suite.
• Prefer a compact cross-product chosen around generated-shell and ownership boundaries.
• Keep vfox-shiv end-to-end install isolation in vfox-shiv; shiv should own generator behavior.
• A failure should preserve the generated shim, paths, environment, and exact invocation as CI artifacts or diagnostic output.

Acceptance criteria

• A repeatable command runs the generated-shim stress matrix locally.
• The matrix runs on Linux and macOS CI.
• It covers hostile valid paths, cache hit/miss/conflict, and same-name version isolation.
• Generated artifacts receive syntax and real execution checks.
• Failures identify the exact case and enough state to reproduce it.