v1.3.1

Performance

• Replace character-by-character Python loop in text newline decoding with C-speed string operations (str.find, in, str.replace), with a fast path that skips translation entirely when no \r is present.
• Eliminate per-line string slice copies in readline() and async for iteration by searching directly in the text buffer with an offset instead of creating a slice copy.
• Track already-scanned buffer positions in readline() / __anext__() to avoid O(n²) re-scanning when lines span multiple chunks.
• Dedicated read(-1) fast path that bypasses per-chunk buffer append/consume overhead, performing a single binary read and decode instead.
• Use memoryview for zero-copy buffer slicing in binary read(), read1(), and readline(), eliminating an intermediate bytearray allocation per call.
• Add binary readline() fast path for the common case where the line fits in the current buffer, avoiding list allocation and b"".join().
• Add binary read(size) early exit when the buffer already satisfies the request.
• Make text _capture_buffer_origin() synchronous by accessing binary position directly, eliminating coroutine creation overhead per chunk.
• Inline small helper methods in hot paths to eliminate method call overhead.
• Pre-compute _universal_newlines flag at init to replace per-chunk set membership tests.
• Defer text buffer compaction until dead space exceeds a threshold, matching the binary layer's strategy.

Benchmark improvements vs 1.3.0

Benchmark	1.3.0	1.3.1	Improvement
Text I/O throughput	17.4 MB/s	37.1 MB/s	2.1x
readline() 10K lines	17.3ms	6.3ms	2.8x
Line iteration 10K	18.2ms	6.5ms	2.8x
JSONL processing	0.78x vs gzip	1.4x vs gzip	Now beats stdlib