Add Python code injection system for remote execution

[Copilot]

Implements AST-based code injection system from codecarbon/pull/1030 to enable dynamic modification of Python scripts before remote execution.

Changes

• Injector class (src/mcp_remote_run/injector.py)

  • Variable injection with type coercion (str, int, float, bool, None)
  • Dependency injection via os.system("pip install ...")
  • Function body replacement via AST rewriting
  • Temporary file management with context manager support
  • Method chaining API

• Package structure

  • pyproject.toml with libcst dependency
  • Test suite (14 tests, 90% coverage)
  • Usage examples

Example

from mcp_remote_run import Injector

code = """
def train_model():
    pass
"""

with Injector(code=code) as injector:
    injector.inject_variables({
        "epochs": 10,
        "batch_size": 32
    }).add_dependency([
        "codecarbon",
        "torch"
    ]).inject_function("""
from codecarbon import EmissionsTracker
tracker = EmissionsTracker()
tracker.start()
# training code here
tracker.stop()
""", "train_model")
    
    # Modified script at injector.get_temp_file_path()
    # Ready for remote execution

All operations use libcst for safe AST manipulation. Temporary files auto-cleanup on context exit or explicit destroy().

Original prompt

Use mlco2/codecarbon#1030 as a starting point.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.