max_buffer_size and import byte limits count characters instead of UTF-8 bytes

### Summary

Several limits documented and reported as byte limits use `len(str)`, which counts Unicode code points rather than encoded UTF-8 bytes.

### Affected code

- `src/claude_agent_sdk/_internal/transport/subprocess_cli.py:68-90` — `_LineFramer`
- `src/claude_agent_sdk/_internal/transport/subprocess_cli.py:800-808` — stderr buffering
- `src/claude_agent_sdk/_internal/transport/subprocess_cli.py:970-987` — stdout message guard
- `src/claude_agent_sdk/_internal/session_import.py:128-142` — import batching

### Current behavior

The framer increments `pending_len` with `len(chunk)`, and completed lines are checked with `len(line)`. Error messages describe the result as bytes. Session import similarly says it flushes at 1 MiB of line bytes but increments with `len(line)`.

For multibyte UTF-8 text, the actual byte count can be significantly larger. A string of emoji, for example, occupies about four UTF-8 bytes per Python character.

### Why this matters

- `max_buffer_size` does not enforce the promised wire/memory boundary.
- A nominal 1 MiB configuration can accept roughly 4 MiB of UTF-8 content before rejection.
- stderr partial-line buffers and SessionStore append batches can grow well beyond their intended byte thresholds.
- Operators relying on these settings for memory protection receive misleading error measurements.

### Expected behavior

A setting and error described as bytes should count encoded bytes consistently.

### Possible fix

The most robust approach is to frame raw bytes and decode complete lines afterward. Alternatively, maintain an exact incremental UTF-8 byte count alongside text chunks and use `len(line.encode("utf-8"))` for completed lines.

Tests should use multibyte input immediately below and above the configured byte boundary, including chunk splits across Unicode content.

### Environment

- Repository revision: current `main` audit at SDK version 0.2.128
- Bundled CLI version: 2.1.220
- Python test suite: 1,291 passed, 5 skipped
- Ruff and mypy: clean

I searched the existing issues and pull requests using the affected symbols and behavior before filing this.