Add extract_screen_text — active-display OCR for the Altic MCP server

# Plan: Add extract_screen_text MCP tool + update the skill

## Context
The repo (altic-mcp, a Python 3.13 / FastMCP server for macOS automation) can capture
the active display as an image (capture_active_screen) but cannot read the text on screen.
We want a new tool, extract_screen_text, that captures the active display and returns
the visible text via macOS Vision OCR — with an optional macOS 27 Foundation Models "visual
summary" mode for higher-level UI understanding. The skill (altic-studio) and README must be
updated so the model knows when/how to use it.

A complete, working reference implementation already exists on the sibling branch
feat/extract-screen-text (10 files, 856 insertions). We are on feat/text-extraction and will
port that work — with two deviations: (1) fix a latent compile bug in the Foundation Models
path, and (2) skip the empty package.json / package-lock.json files (the project is Python/uv,
those are cruft).

## Scope decisions (confirmed with user)
- OCR + visual summary — full parity with the reference branch, including
  include_visual_summary / visual_prompt.
- Skip package.json and package-lock.json.

## Files to create
1. tools/screen_text.py (new, Python wrapper)
   Mirror feat/extract-screen-text:tools/screen_text.py verbatim. Follows existing wrapper
   conventions (compare tools/screenshot.py, tools/clipboard.py):
   - DEFAULT_VISUAL_PROMPT constant.
   - _error(), _json(), _script_path() (uses SCRIPTS_PREFIX from tools/constants.py),
     _default_output_path() (writes to tempfile.gettempdir()/altic-mcp-screen-text).
   - extract_screen_text(output_path, max_chars=20000, include_visual_summary=False,
     visual_prompt=DEFAULT_VISUAL_PROMPT): resolves/creates output path, runs the Swift
     helper via subprocess.run(..., timeout=90), parses the JSON stdout, truncates text to
     max_chars in Python, returns a normalized JSON payload (action, screenshot_path, text,
     length_chars, returned_length_chars, truncated, line_count, average_confidence, plus
     visual_* keys when present).
   - Error handling: Error: ... strings for subprocess failure, invalid JSON, timeout, generic.

2. tools/scripts/extract-screen-text.swift (new, Swift helper)
   Mirror feat/extract-screen-text:..., reusing capture-active-screen.swift display-selection
   logic and adding:
   - recognizeText(in:) using VNRecognizeTextRequest (Vision).
   - Optional FM visual summary gated behind @available(macOS 27.0, *) and
     #if canImport(FoundationModels) && canImport(_Vision_FoundationModels).
   - Emits one JSON object on stdout via JSONEncoder with .withoutEscapingSlashes.
   - CLI args: <output_path> [include_visual_summary] [visual_prompt].
   BUG FIX (deviation): in generateFoundationVisualSummary, change
     let session = LanguageModelSession(model: model, tools: [OCRTool()])
   to
     let session = LanguageModelSession(model: model)
   (OCRTool() is undefined; this breaks compile on a macOS 27 toolchain.)

3. skills/altic-studio/scripts/extract-screen-text.swift (new, mirror)
   Copy the fixed Swift file so the two copies stay identical.

4. tests/test_screen_text.py (new)
   Mirror feat/extract-screen-text:tests/test_screen_text.py — pytest + monkeypatch over
   screen_text.subprocess.run. Covers default args/swift invocation shape, visual-summary
   options passthrough, Python-side truncation, visual-unavailable reporting, subprocess error,
   invalid JSON, and test_server_exposes_extract_screen_text.

## Files to modify
5. server.py — add screen_text to the from tools import (...) block (alpha order, after safari);
   register the tool right after capture_active_screen (before add_screen_glow), mirroring
   feat/extract-screen-text:server.py with Field(default=..., ge=1, le=200000) constraints.

6. skills/altic-studio/SKILL.md — apply the SKILL.md diff from the reference branch:
   - Intro list: add line 6 (MCP screen text mode); amend Swift utility scripts sentence.
   - Mode A capabilities list: add extract-screen-text.swift entry + Swift command template.
   - Mode B (Chrome) tool list: add extract_screen_text.
   - New Mode B2: Screen Text and Visual Understanding (MCP) with tool, args, workflow rules.
   - Operational Rules + Permissions Checklist additions.

7. skills/altic-studio/scripts/README.md — add the example invocation line.

8. README.md — feature bullet, skill listing, Screen Recording / macOS 27 permission updates,
   "Manual Smoke Tests For Screen Text Tools" section.

## Not doing
- No new Python dependencies (Vision/FoundationModels are native; called via Swift subprocess).
- Skip package.json / package-lock.json.

## Verification
1. uv run pytest tests/test_screen_text.py -v ; uv run pytest -q (full suite).
2. Server registration via test_server_exposes_extract_screen_text; optionally import server.
3. swift -typecheck tools/scripts/extract-screen-text.swift (OCRTool fix prevents known error).
4. End-to-end manual on macOS w/ Screen Recording: OCR-only, FM summary on macOS 27, and the
   FM-unavailable fallback; plus a direct script invocation.