Split exact-match NameRes lookups into their own log-analysis notebook

[gaurav]

log-analysis/nameres/analyze_nameres_logs.py was narrowed to autocomplete-only in #107. Exact-match (autocomplete=False) lookups now have no home.

They were split out because mixing the two made every aggregate a weighted average of two populations that behave nothing alike. Measured over a 16,461-row mixed export:

mode	n	p50 Solr wait	p95	p99	max
autocomplete	504	48.5 ms	2,803 ms	263,722 ms	321,254 ms
exact	15,957	14.4 ms	42.7 ms	486 ms	29,830 ms

Exact lookups are ~97% of traffic and the far healthier population, but they are what most NameRes callers actually use, so they still want characterizing before the ElasticSearch comparison.

Suggested shape:

• A sibling marimo notebook over the general data/log-analysis/nameres-*.json exports (the ones the autocomplete notebook's glob deliberately excludes).
• Reuse the existing parser rather than copying it. That probably means promoting QueryLogEntry / parse_record / LOOKUP_LINE_RE out of the notebook cell into src/babel_validation/, at which point tests/log_analysis/test_nameres_log_notebook.py can import them directly instead of going through Cell.run().
• Typing chains do not apply here; the interesting splits are result limit, filter usage, and query length.

Note that the general exports are capped 10,000-row CloudWatch samples, not complete populations — unlike the autocomplete-only export, whose filter is selective enough to fit under the cap.