Live-retrieval: fix file upload crash caused by output path conflict between dispatcher and post-processor

[AZOGOAT]

Summary

Fixes a bug where uploading a file via the live-retrieval API (/v1/files, /v1/files/bulk, /v1/files/{fileId}) crashes with IsADirectoryError.

Bug

When uploading a file via the /v1/files API, the process_files_default function uses the same path (./tmp/my_docs) for two different purposes:

1. The Dispatcher creates it as a folder to store intermediate processing results (images, JSONL files)
2. The post-processing pipeline tries to open it as a file to save chunked output

Since the folder already exists when the post-processor runs, it crashes with IsADirectoryError: [Errno 21] Is a directory: './tmp/my_docs'. The file gets saved to ./uploads/ but is never indexed, so retrieval returns nothing.

Error logs

[INDEX API 🗂️] Results saved to ./tmp/my_docs/processors/PDFProcessor/results.jsonl
[INDEX API 🗂️] PP Pipeline:
[INDEX API 🗂️] > 1. 🦛 Chunker
🦛 Chunker: 100%|██████████| 1/1 [00:00<00:00, 55.85it/s]
[INDEX API 🗂️] Failed to save samples to ./tmp/my_docs: [Errno 21] Is a directory: './tmp/my_docs'
[INDEX API 🗂️] Error uploading file: [Errno 21] Is a directory: './tmp/my_docs'

File "/app/src/mmore/utils.py", line 240, in process_files_default
chunked = pipeline(raw_documents)
File "/app/src/mmore/process/post_processor/pipeline.py", line 68, in call
return self.run(samples)
File "/app/src/mmore/process/post_processor/pipeline.py", line 98, in run
save_samples(samples, self.output_config.output_path)
File "/app/src/mmore/process/utils.py", line 101, in save_samples
with open(path, mode) as f:
IsADirectoryError: [Errno 21] Is a directory: './tmp/my_docs'

Fix

Append /results.jsonl to the post-processor output path so it writes a file inside the existing directory instead of trying to overwrite it.

[fabnemEPFL]

still needed? @AZOGOAT