fix(buffers): make byte limits Unicode-correct

[GautamSharma99]

Summary

Make every limit documented as bytes enforce the actual UTF-8 encoded size instead of Python Unicode character counts.

• track pending stdout/stderr line size incrementally in UTF-8 bytes
• validate completed stdout messages using their encoded byte length
• batch imported JSONL entries using encoded line bytes
• add multibyte boundary regressions for stdout, stderr, and session import

Problem

len(str) counts Unicode code points, not bytes. That made max_buffer_size and MAX_PENDING_BYTES substantially more permissive for multibyte content. For example, emoji generally require four UTF-8 bytes per Python character, so a nominal 1 MiB byte limit could allow several MiB of encoded data.

The mismatch affected three related protections:

1. _LineFramer tracked pending stdout and stderr fragments in characters.
2. Complete stdout lines were guarded using character length.
3. Session import batches accumulated JSONL line characters while documenting a byte threshold.

Implementation

The transport still frames decoded text, preserving the existing stream and parsing behavior. _LineFramer now maintains pending_bytes by adding the UTF-8 encoded size of each decoded chunk and recomputing the encoded size of the residual tail after newline framing. Complete stdout lines use the same UTF-8 measurement before parsing.

Session import now adds len(line.encode("utf-8")) to its byte accumulator. Batch-size and entry-count semantics remain otherwise unchanged.

This is exact for the decoded UTF-8 text delivered by TextReceiveStream and avoids changing transport decoding or line-framing contracts.

Tests

Added regressions proving that:

• a multibyte stdout message split across chunks is accepted exactly at its encoded byte size and rejected one byte below it
• a newline-free multibyte stderr fragment flushes when encoded bytes exceed max_buffer_size, even when its character count does not
• Unicode JSONL imports flush at MAX_PENDING_BYTES based on encoded bytes and preserve entry ordering

Validation

• uv run --extra dev pytest -q — 1294 passed, 5 skipped
• uv run --extra dev ruff check src tests — passed
• uv run --extra dev ruff format --check src tests — passed
• uv run --extra dev mypy src — passed

Fixes #1165