Fix CLI error double-wrapping, config read side effect, and stale flag docs

[betomoedano]

Summary

Routine bug-scan pass over the codebase. Fixes several small but real bugs; no behavior additions.

Updated after main merged the feature-graphic command (#30): conflicts resolved, and the same two bug patterns found in the icon command were also present in the new feature-graphic command, so they're fixed there too.

Bug fixes

• Double-wrapped error messages (src/commands/icon.ts, src/commands/feature-graphic.ts): every this.error(...) inside run() throws an oclif CLIError that was caught by the outer catch and re-wrapped as Failed to generate icon: <original red message>. For example, snapai icon -p x --model banana -n 2 printed a nested, double-styled message. The catch now re-throws CLIErrors as-is and only wraps unexpected errors (API failures, save failures).

  Before:

  ›   Error: Failed to generate icon: Banana normal only supports -n 1
  

  After:

  ›   Error: Banana normal only supports -n 1
  

• Config file created on read (src/services/config.ts): getConfig() called fs.ensureFile(), so read-only commands like snapai config --show (or any generation using env-var keys) created an empty ~/.snapai/config.json as a side effect. Reads now check pathExists instead.

• Preview/generation validation mismatch (src/commands/icon.ts, src/commands/feature-graphic.ts): --prompt-only skipped validation that generation mode performs. For icon, --model banana-2 -q high --prompt-only previewed fine but failed when actually generating. For feature-graphic, the preview skipped all provider-specific checks (-n limits and --quality resolution). Previews now run the same validation.

• Broken npm script (package.json): removed test:icon-batch, which points at scripts/test-icon-batch.sh — neither the script nor the scripts/ directory exists in the repo.

• Missing webpack externals (webpack.config.js): @google/genai and mime were missing from externals, so pnpm run bundle inlined them while every other runtime dependency stayed external.

Cleanup

• Removed dead ternary rawPrompt ? prompt : prompt in src/services/openai.ts.
• Fixed stale (GPT-Image-1 only) help text on --background / --output-format / --moderation; these apply to all OpenAI models (with a note that gpt-image-2 rejects transparent).
• Fixed a misleading comment in the config command (no-flag invocation shows config, not help).

Verification

• Merged latest main (including feat: Add feature-graphic command for Google Play Store banners #30) into this branch; pnpm run build, pnpm run lint, and pnpm run bundle all pass.
• Manually verified with the built CLI:
  • icon --model banana -n 2 and fg --model banana -n 3 --prompt-only → single clean error (no double wrapping); runtime errors (e.g. missing API key) still get the Failed to generate ...: prefix.
  • config --show with no ~/.snapai → no file/directory created; config --openai-api-key ... still persists correctly.
  • --model banana-2 -q high --prompt-only now fails the same way generation mode does; normal --prompt-only previews (icon and feature-graphic, incl. --app-name/--style) unchanged.

Related open work (not duplicated here)

• PR Fix "Premature close" error on large image responses (gpt-image-2) #33 — "Premature close" fix for large gpt-image-2 responses
• PR Add baseURL configuration for OpenAI API #25 — baseURL configuration