fix: strip scripts: from extension command frontmatter in command mode

[chelsealong]

Description

Fixes #4554.

CommandRegistrar.register_commands() renders extension commands for
non-SKILL.md agents by consuming the source scripts: frontmatter
block for {SCRIPT} substitution via resolve_skill_placeholders().
In the markdown branch (e.g. Copilot's .github/agents/*.agent.md
commands-mode layout), the now-consumed key was never removed from the
frontmatter dict before render_markdown_command() serialized the
full dict as the final file's frontmatter. IntegrationBase.process_template()
— the core template render path — already strips scripts: as an
explicit step, so the two render paths disagreed and a build-time key
leaked into agent-visible output, exactly as described in the issue.

The fix pops scripts from frontmatter right after the
resolve_skill_placeholders() call in the markdown branch of
register_commands(), once it's no longer needed. The pop is scoped
to extension_id is not None (i.e. extension-owned commands only)
because register_commands() is also shared by the preset system,
where an existing test (test_register_commands_inherits_scripts_from_core)
intentionally asserts that a strategy: wrap preset command retains
the merged scripts: key in this same render step. That's a separate,
pre-existing behavior outside the scope of this issue, so it's left
untouched.

The toml and yaml branches are not affected by this leak and need
no pop: render_toml_command() only ever reads the description key
off frontmatter, and render_yaml_command() only ever reads
title/description — neither serializes the full frontmatter dict,
so there is nothing for a scripts: key to leak into. (An earlier
version of this PR added the same pop to those two branches as
a defensive copy-paste; removed after confirming with a standalone
repro that it was dead code — neither branch's output ever contains a
scripts: key, fix or no fix.)

Skills-mode extension rendering (_register_extension_skills /
register_skill_command) was already unaffected — those paths build a
fresh SKILL.md frontmatter dict from scratch and never copy scripts:
into it — so no change was needed there.

Testing

Added test_command_mode_registration_strips_scripts_key in
tests/test_extensions.py, which installs an extension command
declaring scripts: and registers it for the copilot agent (command
mode, .agent.md output) — mirroring the issue's reproduction.

Confirmed the test fails without the fix:

$ git checkout HEAD~1 -- src/specify_cli/agents.py  # (before the fix)
$ uv run pytest tests/test_extensions.py -k test_command_mode_registration_strips_scripts_key -q
FAILED tests/test_extensions.py::TestCommandRegistrar::test_command_mode_registration_strips_scripts_key
  assert 'scripts:' not in '---\ndescri..."$ARGUMENTS"'
  'scripts:' is contained here:
    ---
    description: Scripted command
    scripts:
      sh: .specify/scripts/bash/setup-plan.sh --json "{ARGS}"
      ps: .specify/scripts/powershell/setup-plan.ps1 -Json
    ---

And passes with the fix applied:

$ uv run pytest tests/test_extensions.py -k test_command_mode_registration_strips_scripts_key -q
1 passed in 0.13s

Full test_extensions.py + test_presets.py suite (unmodified elsewhere):

$ uv run pytest tests/test_extensions.py tests/test_presets.py -q
1214 passed in 14.82s

AI Disclosure

This PR was written primarily by an autonomous AI coding agent (Claude
Code, model Claude Sonnet 5), including the investigation, the code
change, and the regression test. I reviewed the diff and the test
output before opening this PR.