Add Client.resolve_activities, concurrent batch name resolution

[ccomb]

Why

Scripts that need process ids for a list of names keep reinventing the same two workarounds: downloading the entire database (limit=200000) to build a name→process_id dict, or hand-rolled thread pools issuing one search_activities per name.

What

Client.resolve_activities(names, *, by="name"|"product", geo=, exact=True, limit=5, workers=8) resolves a batch of names concurrently and returns {name: [matches]}:

• The mapping is total: a name that doesn't resolve maps to [] instead of disappearing — misses stay visible.
• One match = unambiguous; several = ambiguous (same name across geographies), disambiguate with geo= or inspect the candidates.
• exact=False returns relevance-ranked candidates, best first.
• Duplicated input names are searched once; errors propagate instead of degrading to empty results.

With exact=True the client re-checks equality on the returned matches (same casefold rule as the engine): engines whose product filter ignores exact= return substring matches, and a near-miss must never resolve silently. A separate engine PR fixes that filter at the source.

Verification

• 10 offline orchestration tests (mapping totality, dedup, argument passthrough, exactness guard, fail-loud).
• End-to-end against a local engine build: exact hit, miss, fuzzy ranking, by="product", and a deduplicated batch.
• README API reference regenerated (drift test green).