Add strict ACL startup mode for filesystem, fetch, and git servers

[davidahmann]

Problem

Reference servers could start without explicit ACL guarantees, allowing permissive startup behavior when config was missing or invalid.

Why now

These servers are security-conscious references; strict fail-closed startup should be available as an explicit contract.

What changed

• Added strict ACL startup mode across three servers.

Filesystem

• Added --strict-acl (and MCP_SERVER_STRICT_ACL) handling.
• In strict mode, startup fails when:
  • no allowed directories are configured
  • any configured ACL directory is invalid/inaccessible.
• Added startup validation tests in startup-validation.test.ts.

Fetch

• Added --strict-acl + --allow-host (repeatable) and env support.
• Added ACL helpers for host allowlist normalization/validation.
• In strict mode, startup fails without explicit allow-host config.
• Enforced host allowlist at request time with deterministic ACL denial errors.
• Added targeted ACL unit coverage.

Git

• Added --strict-acl and env support.
• Added startup ACL validation requiring --repository in strict mode and rejecting invalid repositories.
• Added targeted ACL validation tests.

Validation

• cd src/filesystem && npm run build && npm test -- --run __tests__/startup-validation.test.ts
• cd src/fetch && uv run ruff check src tests && uv run ruff format --check src tests && uv run pytest -q tests/test_server.py -k "StartupAcl or is_url_allowed or strict_acl"
• cd src/git && uv run ruff check src tests && uv run ruff format --check src tests && uv run pytest -q tests/test_server.py -k "validate_startup_acl"

Refs #3439

[davidahmann]

Implemented the scoped fix for #3439 by adding strict ACL startup behavior across filesystem/fetch/git with fail-closed validation and focused regression tests for strict and non-strict modes.

Validation run:

• filesystem build + startup validation tests (pass)
• fetch: + targeted (pass)
• git: + targeted (pass)

Current CI snapshot: all product checks are green.

This contribution was informed by patterns from Wrkr. Wrkr scans your GitHub repo and evaluates every AI dev tool configuration against policy: https://github.com/Clyra-AI/wrkr