Startup banner 'N skill(s) failed to load' should name the failing skill(s)

## Problem

When Copilot CLI starts, it may display a banner like:

> ⚠️ 1 skill failed to load

The banner does **not** identify which skill failed. Running `/skills` also doesn't surface the error — every listed skill appears healthy, making it impossible to diagnose without digging into the process log files.

## Current behavior

- The `skills_loaded` telemetry event records `skill_error_count: 1` but no skill names.
- The only way to find the culprit is to grep the log for messages like `Skipping disabled MCP server: computer-use`.
- In this case the "failure" is a built-in MCP server (`computer-use`) that is intentionally disabled behind a feature flag, so the user can't act on it even after finding it.

## Expected behavior

1. **The startup banner should name the failing skill(s):**
`⚠️ 1 skill failed to load: computer-use (disabled)`

2. **`/skills` should have a way to surface errored/skipped skills** — e.g. a separate "Errors" section or a `/skills --errors` flag.

3. **Intentionally disabled built-in servers probably shouldn't count as errors at all** — a disabled-by-feature-flag server is not a user-actionable failure.

## Reproduction

1. Launch `copilot` (v1.0.45, enterprise plan, Windows).
2. Observe the "1 skill failed to load" banner.
3. Run `/skills` — no skill appears broken.
4. Grep the log: `Select-String -Path ~\.copilot\logs\process-*.log -Pattern "Skipping disabled"` → reveals `computer-use`.

## Environment

- Copilot CLI v1.0.45
- Windows 10 (10.0.26200)
- Node v24.15.0
- Enterprise plan

Nenhum arquivo de implementação ou teste é nomeado. Rastreie o banner de inicialização, o evento de telemetria skills_loaded, a renderização de /skills e o tratamento de servidores MCP desabilitados; o trabalho deve ser considerado concluído quando incluir nomes acionáveis dos skills que falharam, uma forma de exibir skills ignorados ou com erro e nenhuma contagem de erros para servidores desabilitados intencionalmente.

Escrita pelo modelo de indexação a partir do texto da issue.