Improve reporting import and report generation

[ftshijt]

Summary

• Make optional metric modules lazy-loaded so reporting, visualization, and metric discovery can run without importing model-backed dependencies.
• Add metadata-backed reporting categories/directions and wire versa-visualize to source-discovered metric metadata.
• Add versa-score --report report generation for utterance-level scoring outputs.
• Add offline-focused tests for lightweight imports, registry alias metadata, and scorer report export.

Validation

• /opt/homebrew/Caskroom/miniforge/base/envs/versa-dev/bin/python -m pytest test/test_reporting.py test/test_metrics/test_definition.py -q
• /opt/homebrew/Caskroom/miniforge/base/envs/versa-dev/bin/python versa/bin/visualize.py demo/reporting_example_results.jsonl --out /tmp/versa-report-smoke.html
• /opt/homebrew/Caskroom/miniforge/base/envs/versa-dev/bin/scorer.py --score_config egs/separate_metrics/pesq.yaml --gt test/test_samples/test1.scp --pred test/test_samples/test2.scp --output_file /tmp/versa-score-report-results.jsonl --io soundfile --report /tmp/versa-score-report.html
• /opt/homebrew/Caskroom/miniforge/base/envs/versa-dev/bin/python versa/bin/scorer.py --list-metrics >/tmp/versa-list-metrics.txt

Upstream target intended for final review: wavlab-speech/versa:main.