perf(docs): prefetch the mirror bundle before a first search

[m1ngsama]

A first docs search downloaded all 247 documents one by one: 40 s and 248 requests on this network.

• Search prefetches the whole library from docs.nbtca.space/docs-api/bundle.json through @nbtca/docs 0.3.4 when more than a handful of documents are uncached, then loads only what's still missing.
• A cold search drops from 40 s / 248 requests to 6.6 s / 2 requests; warm searches are unchanged.
• The progress counter starts at the number already available, so it no longer climbs through documents that arrived in the bundle.