fix: prevent pandas_kwargs mutation during text file compression resolution

[hsusul]

Description

This PR fixes a bug where _get_read_details and _get_write_details mutated the passed pandas_kwargs dictionary in place when resolving file compression (e.g. setting pandas_kwargs["compression"] = infer_compression(...)).

Symptom & Root Cause

When reading a list of S3 paths or a dataset containing files with different compression extensions (such as ["s3://bucket/file1.csv.gz", "s3://bucket/file2.csv"] or ["s3://bucket/file1.json.gz", "s3://bucket/file2.json"]), _get_read_details mutated the shared pandas_kwargs dictionary during the processing of the first file (e.g., setting "compression": "gzip").

Because pandas_kwargs was mutated in place:

1. pandas_kwargs.get("compression", "infer") returned "gzip" for subsequent files rather than "infer".
2. Compression inference was skipped for all remaining files in paths.
3. Reading mixed compressed/uncompressed files failed with errors such as gzip.BadGzipFile: Not a gzipped file, UnicodeDecodeError, or produced corrupted data.

Solution

• Updated _get_read_details and _get_write_details to operate on a copy of pandas_kwargs per-file, preserving pandas_kwargs across multiple files so each file infers its own compression setting independently.
• Added unit regression tests in tests/unit/test_moto.py covering read_csv and read_json with mixed compression paths and chunked reads.

Validation Results

• pytest tests/unit/test_moto.py: All 47 tests passed.
• git diff --check: Clean (0 errors).
• ruff check: All checks passed.
• ruff format --check: All files formatted.
• mypy: Success (0 errors).