feat: Add expected_outputs for conditional output computation

[bigcat88]

This PR adds the ability for nodes to know at runtime which of their outputs are actually connected downstream. This enables nodes to skip computing expensive outputs that won't be used.

Motivation

Some nodes (especially API nodes) produce multiple outputs in different formats (e.g., GLB, FBX, OBJ for 3D models). Currently, all outputs are computed even if only one is connected. This wastes time and bandwidth downloading unused files.

• During execution, expected_outputs is computed by scanning which outputs have downstream connections
• Nodes access this via get_executing_context().expected_outputs
• Outputs NOT in the set are definitely unused and safe to skip
• Outputs IN the set might be used

Example of usage:

from comfy_execution.utils import is_output_needed

...


# Output indices: 0=GLB, 1=FBX
glb_file = None
fbx_file = None
if is_output_needed(0):
    glb_file = await download_url_to_file_3d(result.model_urls.glb, "glb", task_id=task_id)
if is_output_needed(1):
    fbx_file = await download_url_to_file_3d(result.model_urls.fbx, "fbx", task_id=task_id)
return IO.NodeOutput(glb_file, fbx_file)     

This feature may also be useful for non-API nodes as nodes can now even change their behavior based on unconnected outputs.

[guill]

nit: Instead of get_expected_outputs_for_node (and using the word expected in general), what do you think of get_possible_outputs_for_node or something like that?

[bigcat88]

"Possible outputs" sounds like it describes the node capability - which output slots it could produce. A node with 3 output slots always has 3 possible outputs, regardless of connectivity.