Exact counts, uniform random, faster reads

This release sharpens the pieces you build quizzes on: accurate totals when you ask for them, unbiased random questions, and lower latency on the most common reads.

Added

Exact match counts on demand. Add ?count=exact to /api/v1/questions and the response includes meta.total — the precise number of questions matching your filters, computed in milliseconds. Use it when you want to show "N results"; skip it when you're just paging.

Changed

Pagination is cursor-first by default. Without ?count, responses no longer include a total — page forward with the cursor and _links.next (the pattern large APIs use). Ask for ?count=exact only when you need the number. The field is meta.total.
Random sampling is now uniform. /api/v1/questions/random returns an evenly distributed, unbiased sample across the whole matching set — previously sampling leaned toward certain slices of the catalog. Same call, better spread.
Faster reads. Listing, category-filtered queries, random, and single-question lookups all return quicker — typical reads are in the tens of milliseconds when warm. See /docs/performance for the current numbers.
Per-question totals stay exact in discovery. /api/v1/tags, /subcategories, /topics, and /regions already return an exact count per item — the place to look when you want "how many questions are in X" without scanning.
TypeScript SDK. @quizbase/client is now 0.9.0 — meta.total (replacing the old estimate field) and count: 'exact' | 'none' typings.

Removed

The approximate count (?count=estimate). The estimate could be far off on narrow filter combinations, so we replaced it with the exact opt-in count above. If you were sending ?count=estimate, switch to ?count=exact for an accurate total or drop the parameter to page by cursor.