Fix Sequential double execution bug (#10393)

[DhyeyMavani2003]

Description

Fixes #10393 - Sequential initialization causing program to run twice when called from a script with a valid Python identifier name.

Problem

When torch_geometric.nn.Sequential was initialized from a script with a valid Python identifier name (e.g., foo.py), the entire program would execute twice. This only occurred when the filename was a syntactically valid Python function name.

Example

# foo.py
import torch_geometric.nn as pyg

print("before")
pyg.Sequential('x', [(lambda x: x, 'x -> y')])
print("after")

Running python foo.py would output:

before
before
after
after

Root Cause

The jinja template used by Sequential was importing the caller module with from <module> import *. When the caller was the __main__ script (e.g., foo.py), this caused Python to import and execute the module again, resulting in double execution.

Solution

Modified _set_jittable_template method to filter out the caller module from imports if:

• The caller module is __main__ (the script being executed)
• The caller module is not yet in sys.modules (would be imported for the first time)

This prevents circular imports and double execution while maintaining functionality for legitimate module imports.

Changes

• Modified torch_geometric/nn/sequential.py to add filtering logic before passing modules to the jinja template
• Added test case test_sequential_no_double_execution to prevent regression

Testing

• ✅ Reproduced the original bug and verified the fix resolves it
• ✅ All existing tests in test/nn/test_sequential.py pass (7/7)
• ✅ New test validates no double execution occurs during initialization
• ✅ Tested with both valid Python identifier filenames and hyphenated filenames

Checklist

• I have read the contributing guidelines
• I have added tests that prove my fix is effective
• I have updated the documentation (if applicable)
• All new and existing tests passed