fix(extensions): reject duplicate provides.templates/scripts names

[chelsealong]

Summary

Follow-up to #4012 (#4010), which merged before the last round of Copilot review feedback could be addressed. This carries forward the two outstanding items that are still valid against main:

• provides.templates/provides.scripts entries with a duplicate name within the same section are now rejected with a ValidationError. The resolver (PresetResolver._extension_manifest_declared_template) returns the first entry matching a name, so a later duplicate was silently unreachable while still being exposed by ExtensionManifest.templates/.scripts — this closes that gap at manifest-validation time instead. Added parametrized regression tests for both templates and scripts.
• extensions/EXTENSION-DEVELOPMENT-GUIDE.md's provides section now distinguishes its own sub-fields (commands/templates/scripts) from hooks/events, which are top-level manifest fields read from the manifest root, not nested under provides — the prior wording could lead an author to indent them under provides, where they wouldn't satisfy validation.

The other two items from that last review round were doc-only / description-only and don't apply here: #4012's PR description has already been corrected in place, and the AGENTS.md acceptance-criterion item was intentionally left alone since that file documents only the AI-agent integration subsystem, not the extension system (which has its own docs, already updated).

Testing

New parametrized test test_provides_entry_duplicate_name_rejected in tests/test_extensions.py::TestExtensionManifestTemplatesAndScripts covers both templates and scripts. Confirmed it fails with DID NOT RAISE ValidationError against the pre-fix source (git checkout HEAD~1 -- src/specify_cli/extensions/__init__.py), then passes after restoring the fix.

$ .venv/bin/python -m pytest tests/test_extensions.py tests/test_presets.py -q
====================== 1110 passed, 2 warnings in 11.30s ======================

$ uvx ruff@0.15.0 check src tests
All checks passed!

AI Disclosure

• I did not use AI assistance for this contribution
• I did use AI assistance (describe below)

Implemented autonomously by Claude Code (model: Claude Sonnet 5) under human direction, in response to outstanding automated review feedback on #4012 that had not been incorporated before that PR merged.