feat(scripts): add opt-in parallel execution to convert & install

[CagesThrottleUs]

Summary

Adds --parallel and --jobs N to scripts/convert.sh and scripts/install.sh so users can run tool-level conversion/installation in parallel. Serial remains the default to avoid breaking existing usage.

• convert.sh: When --parallel and tool is all, runs antigravity, gemini-cli, opencode, cursor, openclaw, qwen via xargs -P; runs aider and windsurf serially (shared temp files). Output is buffered per tool so each tool’s output stays together.
• install.sh: When --parallel, runs each selected tool’s install in parallel with output buffered per tool. Works with both interactive and non-interactive modes.
• Job count: Defaults to nproc (Linux), sysctl -n hw.ncpu (macOS), or 4. Overridable with --jobs N.
• Progress: Optional tqdm-style progress bar [=======> ] 3/8 in both scripts (serial and parallel).
• README: Documents parallel usage and recommends opt-in parallel for faster runs.

Testing

• Manual: Ran ./scripts/convert.sh (serial) and ./scripts/convert.sh --parallel --jobs 2; compared output and checked integrations.
• Output parity: Generated outputs to separate dirs (./scripts/convert.sh --out ./TEST-SERIAL, parallel run with --out ./TEST-PARALLEL) and compared with:
  • diff -rq TEST-PARALLEL TEST-SERIAL (no output = identical file sets and contents).
  • diff -r TEST-SERIAL TEST-PARALLEL (no output = byte-identical files).
• Install: Ran ./scripts/install.sh --no-interactive and ./scripts/install.sh --no-interactive --parallel for selected tools; confirmed installs and output grouping.

Testing limitations

• No automated CI for the parallel path; all checks were manual.
• Parallel vs serial comparison is by diffing output directories and logs; no dedicated test suite.
• Job count defaults depend on nproc/sysctl availability (Linux/macOS); other platforms may fall back to 4.
• Test scripts (test-convert-parallel.sh / test-install-parallel.sh) if present are for local comparison only and are not run in CI.

AI and human roles

• Human: Design choices (minimal change, tool-level parallel only, opt-in --parallel, use of xargs and nproc/sysctl); validation of behavior; testing (serial vs parallel, diff of TEST-SERIAL vs TEST-PARALLEL); decision to adopt or reject suggested changes.
• AI (Cursor): Implementation of --parallel/--jobs, progress bar, and buffered output; locating serial execution points; suggesting structure for parallel runs and README wording. AI did not run tests; human ran commands and confirmed results.

[CagesThrottleUs]

This should solve the issue - #187