perf: defer Traceback import in logging.py

[KRRT7]

Summary

• Defer from .traceback import Traceback from module-level to inside emit(), only when rich_tracebacks is enabled
• Avoids loading the traceback → syntax → pygments chain (~20ms) for the default RichHandler usage

Benchmarks (Standard_D2s_v5, Python 3.13, hyperfine)

Module	Before	After	Delta
RichHandler	90.8 ms	82.4 ms	-8.4 ms
Console → RichHandler gap	22.4 ms	3.0 ms	-19.4 ms

The gap between Console and RichHandler (which isolates the Traceback import cost) collapsed from ~22ms to ~3ms.

Reproduce locally

# Quick single-branch test:
pip install -e .
python benchmarks/bench_import.py

# Compare branches with hyperfine (recommended):
python benchmarks/bench_import.py --compare master perf/lazy-traceback-import

The --compare flag uses hyperfine's native --parameter-list to measure both branches in a single invocation, producing a statistically rigorous relative comparison:

# What it runs under the hood:
hyperfine \
  -L branch master,perf/lazy-traceback-import \
  --setup 'git checkout {branch} && python -m pip install -e . -q' \
  -n '{branch}' \
  'python -c "from rich.logging import RichHandler"'

Checks

• black --check passes
• mypy -p rich --no-incremental passes (0 issues)
• pytest tests/ passes (952 passed, 25 skipped)
• CHANGELOG.md updated
• CONTRIBUTORS.md updated

[KRRT7]

Superseded by the consolidated stacked PRs #12 + #13.