feat: validate connections and inspect values in editor

[philippjfr]

Description

I integrated ReactFlow's drag-time connection validation and port/edge value inspection into the FlowDash editor. A side-effect-free DataflowGraph.validate_connection checks declared port direction, duplicate connections, occupied single inputs, cycles, and scalar type compatibility before edge creation; the same rules supply rejection reasons during a drag. List inputs still accept multiple scalar sources. Hovering a port or edge now previews its current value by default, without an extra Paper inside the popup; popup_trigger, popup_hover_delay, and value_repr remain configurable.

The ReactFlow dependency is raised to >=0.5.0b0 in both pyproject.toml and pixi.toml. Tests: PYTHON_GIL=1 pixi run test (330 passed, 2 skipped), PYTHON_GIL=1 pixi run -e test-ui pytest tests/ui --ui --browser chromium -n 1 --tb=short -q (2 passed), and pre-commit on edited files.

Before: Invalid connections were created briefly before server rejection; inspecting a value required a click and showed a nested Paper. After: Drag targets show rejection reasons before connection, and hovering an existing port or edge shows the value on one popup surface. To try this locally, run flowdash serve examples/complex_dataflow, open the editor, drag an output toward an incompatible or occupied input, then hover a port or existing edge.

AI Disclosure

Tool & Model: Kilo Code + GPT-6 Sol
Usage: Implemented the editor integration, validation, tests, and PR description with AI assistance; reviewed the diff and ran the listed checks.

• I have tested all AI-generated content in my PR.
• I take responsibility for all AI-generated content in my PR.

Checklist

• Tests added and are passing