Add kernel execution support for custom nodes

[Edwardvaneechoud]

Summary

This PR adds support for executing custom nodes inside Docker containers (kernels) with configurable package dependencies, while maintaining backward compatibility with the existing lazy-plan execution model.

Key Changes

Frontend

• Node Designer: Added kernel execution toggle and required packages configuration UI

  • New "Execute in Kernel" checkbox to enable kernel mode
  • Input field for comma-separated list of required packages
  • Automatic code template switching between process method and kernel code

• Custom Node Component: Implemented kernel selection UI with real-time status monitoring

  • Dropdown to select available kernels with state indicators (idle, executing, starting, stopped, error)
  • Visual indicators for kernel state and missing packages
  • Link to Kernel Manager for managing kernels
  • Warning messages when no kernel is selected or kernel is not in idle state
  • Polling mechanism (5-second intervals) to keep kernel status up-to-date

• Code Editor: Updated to support both execution modes

  • Dynamic header and hints based on execution mode
  • Validation rules adapted for kernel vs. lazy execution

• Constants & Types: Added kernel-specific defaults and type definitions

  • defaultKernelCode template with flowfile.read_input() and flowfile.publish_output()
  • Extended NodeMetadata and NodeUserDefined types with kernel fields

Backend

• Flow Graph: Implemented dual execution paths for custom nodes

  • _add_user_defined_node_lazy(): Standard lazy-plan execution (existing behavior)
  • _add_user_defined_node_kernel(): New kernel-based execution with:
    • Automatic input materialization to Parquet files
    • Code injection of settings as Python variables
    • Kernel communication via ExecuteRequest
    • Output reading from Parquet back into LazyFrame
    • Proper logging and error handling

• Custom Node Base: Extended with kernel configuration fields

  • use_kernel: bool - Enable kernel execution mode
  • required_packages: list[str] - Docker packages to install
  • kernel_code: str - Python code to execute in kernel

• Code Generation: Updated to generate appropriate node classes

  • Kernel mode: Stores kernel_code as triple-quoted string, stubs process() method
  • Lazy mode: Extracts and re-indents process() method body (existing behavior)

• Node Designer API: Enhanced to parse and extract kernel metadata from custom node files

  • Parses use_kernel, kernel_code, and required_packages attributes
  • Routes to appropriate code extraction based on execution mode

Data Models

• Extended UserDefinedNode schema with optional kernel_id field
• Updated CustomNodeSchema interface with kernel configuration fields

Implementation Details

• Kernel execution follows the same pattern as add_python_script: materialize inputs → send code to kernel → read outputs
• Settings are injected as plain Python variables in the kernel code (no self.settings_schema access needed)
• Kernel polling is automatically started when a kernel-backed node is loaded and stopped on component unmount
• Backward compatible: existing custom nodes continue to use lazy execution by default