Harden CLI failure handling and flag wiring

[takeokunn]

Summary

• remove CLI routes that reported success without an implementation
• connect existing doc, type, ABI, demangle, Wasm inspection, and dependency graph paths
• wire --strict-no-alloc and register --format
• propagate installation and CLI failures instead of swallowing them
• add parser, dispatch, invalid-input, and option regression tests

Verification

• local structural checks pass: git diff --check, Lisp parenthesis checks, fake-handler scan, and CLI-source ignore-errors audit
• local CLI test startup is blocked because :CL-CC-BOOTSTRAP is unavailable; CI is required for the test verdict

Merge is intentionally left to the maintainer after review.

[takeokunn]

Independent review follow-up

Review findings and responses:

• Correctness: the default eval and REPL lazy stdlib fallbacks are restored. The watch loop catches only file-error for editor atomic-save renames, and demangle reports a missing name argument. No production-code defect remains.
• Test strength: fallback tests now cover run, eval, and REPL. The dispatch test executes completion, docs, and version with empty input and checks non-empty output. It iterates every other dispatch entry and requires an error or a positive CLI exit code for empty input; file-command coverage includes missing paths and abi-dump/abi-check.
• Explicit no-argument exceptions: repl (interactive input), selfhost (profile workload), symbols (workspace scan), compile-commands (writes generated output), dep-graph (loads the current system), and update (changes dependency state) are listed with reasons and are not invoked by the sweep. This is the explicit exception path allowed by the task requirements.
• AI slop: no unnecessary production abstraction or swallowed failure path was found. The reviewer’s remaining note is limited to the explicitly listed side-effecting no-argument exceptions.

Verification: CI run https://github.com/nerima-lisp/cl-cc/actions/runs/37571674010 completed successfully. The log contained 219 CLI test result lines and 7488 total tests: 0 skipped, 0 todo, 0 failed, 0 errored.