Fix race condition in batch_predict by replacing file-based processing with in-memory DataFrame

[BHUVANA-SANKU24]

Summary

This PR fixes the race condition in the /api/session/batch_predict endpoint by replacing file-based intermediate storage with in-memory processing using pandas DataFrames.

---

Problem

The previous implementation used a temporary CSV file (temp_batch_predict.csv) to pass prediction data. This caused:

• Race conditions under concurrent requests
• File I/O overhead affecting performance
• Potential file conflicts in multi-user scenarios

---

Solution

• Replaced temporary CSV-based workflow with in-memory pandas DataFrame (df_predict)
• Updated predict_new_data_simple to accept DataFrame input instead of file path
• Removed unnecessary file writing logic

---

Additional Improvements

• Removed debug print statements for cleaner production code
• Simplified route handling in main.py by delegating logic to session.py
• Cleaned up redundant and dead code

---

Testing

• Tested /health endpoint
• Tested /api/session/batch_predict using curl
• Verified correct prediction outputs
• Confirmed no file creation and stable behavior across repeated requests

---

Related Issue

Fixes #82

---

Notes

Happy to align with existing approaches if maintainers prefer a different design direction.