fix(python): support buffer and line reads on S3 readers
Reading an S3 CSV line by line with for line in reader or readline() raises ValueError: I/O operation on uninitialized object. Reading into a reusable buffer with readinto() fails the same way, as does hashlib.file_digest(reader, "sha256"). S3BufferedReader manages its own HTTP stream and byte buffer without initializing the inherited BufferedReader.
Override these read operations to use the existing buffer and _fill() method, and report the reader as readable. Line reads stop at a newline or size limit and preserve the remaining bytes without draining the entire object. Reject read-only destinations before consuming data.
Regression tests in test_s3_reader.py use the existing fake HTTP stream and client. They cover partial buffer reads, memoryviews, EOF, empty destinations, hashing, limited line reads across chunks, and incremental CSV line iteration. On main, five regressions fail with the uninitialized-object error; with the fix, all ten tests pass.
This change was prepared with assistance from Codex.
Summary by cubic
Fixes S3BufferedReader so readinto(), readline(), and line iteration work instead of raising ValueError: I/O operation on uninitialized object. The reader now uses its existing buffer and _fill() method, reports itself as readable(), and rejects read-only destinations before consuming data.
Adds five regression tests covering partial buffer reads, memoryviews, EOF, empty destinations, hashing, limited line reads across chunks, and incremental CSV line iteration.