Recursive Markdown Parsing & Path Preservation

[parssky]

This PR introduces recursive Markdown parsing to AutoRAG without breaking existing behaviour. When the parser receives a glob pattern such as **/*.md, it discovers *.md files at any depth, processes them once, and guarantees that original absolute paths are preserved in all downstream artifacts.

Features

autorag/parse.py:
• Added recursive arg to Parser.start_parsing (default False).
• Converts project_dir to Path to avoid .rglob() error.
• When recursive=True:  Collect all matching Markdown files via glob(..., recursive=True). shutil.copy2 each file into a temp flat dir with path‑encoded filenames. Run run_parser once. Rewrite temp paths in Parquet/JSONL outputs back to originals.• Cleans up temp dir on exit.

Tests:
New tests/autorag/test_parser_recursive.py adds coverage for deep folders; all tests now pass.

Lints
Fixed F823 in api/tests/test_app.py; repo‑wide trailing‑whitespace / EOF cleanup by pre‑commit.

Usage

parser = Parser(data_path_glob="./raw_data/**/*.md", project_dir="./parse")
parser.start_parsing("./parse.yaml", recursive=True)

Temp‑dir strategy
AutoRAG’s run_parser still expects a flat *.md glob. Copying files into a temp folder preserves modification time (via shutil.copy2) and avoids macOS multiprocessing issues with symlinks. After parsing, a two‑line _rewrite_output_paths() post‑process restores the real paths.

Backward Compatibility
Default remains non‑recursive; existing integrations are unchanged.
No LangChain modules were edited.

[parssky]

Hi, several tests are failing in CI, but as far as I can tell, they’re not related to my changes. My own tests are passing, and the errors seem to stem from unrelated modules (e.g., llama_cloud_services, recency_filter, time_reranker, etc.). Could you please confirm if these are unrelated to this PR?