Summary

• remove the unsupported itemsPerPage property passed to fetch
• cache Hydra resource parameter discovery per Resource instance
• deduplicate concurrent parameter requests
• cache successful empty results
• evict rejected requests from the cache to allow retries
• document the cache lifetime

Context

Hydra resource parameters are discovered by requesting the collection endpoint
and reading its hydra:search mapping.

The parser previously passed itemsPerPage as a RequestInit property. This
property is not supported by fetch and was silently ignored. In addition,
concurrent calls to getParameters() or repeated calls returning no parameters
could trigger multiple identical collection requests.

The result is now stored in a WeakMap keyed by the Resource instance. This
keeps caches isolated between documentation parses and allows entries to be
garbage-collected with their resource.

Rejected requests are removed from the cache so that a later call can retry.

This change does not remove the initial Hydra discovery request or assume an
API Platform-specific pagination parameter. It prevents repeated discovery
requests in a generic way.

Related to #115.
Related to api-platform/admin#508.

Tests

Added dedicated coverage for:

• successful caching and concurrent request deduplication
• empty parameter results
• retries after a failed request
• cache isolation between resources sharing the same URL

The parser test retains integration coverage for wiring getParameters() and
the resulting fetch request.

Validation:

• 30 tests passing
• TypeScript typecheck
• Oxlint
• Prettier
• git diff --check