feat: MCP Bridge Workbench, just command cleanup, testing, etc. (#24)

* fix: lots of fixes and name refactoring

* feat: Add workbench preferences

* fix: MCP bridge status widget and just command fixes

* fix(tests): Use the correct mesa-glx package

* fix(ci): Add fontconfig to GUI test dependencies

FreeCAD GUI was failing to start with:
"Fontconfig error: Cannot load default config file: No such file"

Added fontconfig and fonts-dejavu-core packages to the GUI test job
dependencies to resolve the font configuration issue.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* refactor(addon): Extract path utilities into shared module

Create path_utils.py module that consolidates duplicated path-finding
logic from commands.py and InitGui.py:
- get_addon_path(): Find addon directory with caching and fallbacks
- get_icon_path(): Get full path to an icon file
- get_icons_dir(): Get path to icons directory
- get_workbench_icon(): Get path to workbench main icon

This removes ~100 lines of duplicated code while preserving the same
behavior including _addon_path_cache and all fallback methods.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(addon): Prevent stale plugin state on startup failure

The StartMCPBridgeCommand.Activated method could leave _mcp_plugin in
a partially initialized state if FreecadMCPPlugin.start() failed after
the plugin was instantiated.

Changes:
- Create plugin in a local variable first
- Only assign to _mcp_plugin after start() succeeds
- Explicitly clear _mcp_plugin and _running_config in exception
  handlers to ensure clean state for subsequent retry attempts

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix: Lot of broad improvements

* fix(ci): Use blocking headless_server.py for GUI tests

The GUI test was using startup_bridge.py which is non-blocking
(designed for interactive use). For CI, even in GUI mode, we need
the blocking headless_server.py that calls run_forever() to keep
FreeCAD running. GUI features are still available since we use
the 'freecad' executable instead of 'freecadcmd'.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* refactor(addon): Rename headless_server.py to blocking_bridge.py

The old name was misleading because:
- It works with both GUI (freecad) and headless (freecadcmd) modes
- The key characteristic is that it BLOCKS with run_forever()

New naming convention clarifies the difference:
- blocking_bridge.py: Starts bridge and blocks (for CI, servers)
- startup_bridge.py: Starts bridge and returns (for interactive GUI)

Updated all references across:
- GitHub workflow (macro-test.yaml)
- Just commands (freecad.just)
- Unit tests (test_addon_structure.py)
- Documentation (5 files)
- CLAUDE.md

Also improved the script to detect GUI mode dynamically using
FreeCAD.GuiUp and display the appropriate status message.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(just): Remove erroneous rm of startup_bridge.py on error

The startup script is now a permanent source file in the repository,
not a generated temporary file. The rm -f would have deleted source
code if FreeCAD wasn't found.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix: General improvements

* fix: Lots of general fixes and only stable to PyPi

* fix: small cleanup

* fix: Small fixes and hopefully fixes the GUI tests

* fix: Add proper library paths for FreeCAD GUI in CI

- Create wrapper scripts instead of symlinks for AppImage binaries
- Set LD_LIBRARY_PATH, QT_PLUGIN_PATH for GUI mode
- Add diagnostic output to identify startup failures

* fix: Use apprun for GUI tests in CI

* fix: Improving Xvfb tests

* fix: GUI tests worlk

* chore: remove invalid --no-splash comments

* fix: ARM64 architecture support and other fixes

* fix: cleanup

* test: just commands test suite

* test: improve just command tests

* fix: more general improvements

* fix: more cleanup

* fix: more updates

* fix: small tweaks

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
 [project]
 name = "freecad-robust-mcp"
 dynamic = ["version"]
-description = "MCP (Model Context Protocol) server for FreeCAD integration with Claude Code and other AI assistants"
+description = "Robust MCP Server for FreeCAD integration with Claude Code and other AI assistants"
 readme = "README.md"
 license = "MIT"
 requires-python = ">=3.11"
@@ -125,6 +125,8 @@ target-version = "py311"
 line-length = 88
 src = ["src", "tests", "macros"]
 extend-include = ["*.FCMacro"]
+# Exclude auto-generated files (hatch-vcs version file)
+exclude = ["src/freecad_mcp/_version.py"]
 
 [tool.ruff.lint]
 select = [
@@ -287,6 +289,15 @@ markers = [
 source = ["src/freecad_mcp"]
 branch = true
 parallel = true
+# Exclude bridge modules from unit test coverage - they require a running
+# FreeCAD instance. These modules are exercised by integration tests which
+# run separately (just testing::integration). To verify bridge coverage:
+#   uv run pytest tests/integration --cov=freecad_mcp.bridge --cov-report=term-missing
+omit = [
+    "src/freecad_mcp/bridge/embedded.py",
+    "src/freecad_mcp/bridge/socket.py",
+    "src/freecad_mcp/bridge/xmlrpc.py",
+]
 
 [tool.coverage.report]
 exclude_lines = [