apifetch 0.2.0

Bug fixes

af_fetch_all() no longer loops forever with af_paginate_none(): an API that cannot be paged is now fetched with a single request.
af_fetch_all() now accepts chunk_size = Inf (previously failed with a “missing value” error), and truncates to total_limit when an API returns more rows than requested.
parse_queries() (and therefore the query argument of af_fetch()) appends to an endpoint that already has a query string instead of producing a second ?; drops NULL and NA values instead of sending a= or a=NA; repeats a key for multi-valued parameters; and errors clearly on an unnamed list.
Integer-like arguments (limit, offset, chunk_size, …) now reject vectors, NULL, NA, -Inf and values beyond the integer range with a clear message, instead of silently becoming NA or failing cryptically.
Token names with accents are sanitized to plain ASCII on every platform (macOS previously produced names such as Sa'ude).
af_list_tokens() matches the service prefix literally rather than as a regular expression, and skips empty variables.

af_store_token() gains overwrite = FALSE, to rotate an expired token.
af_fetch() returns an empty tibble for responses without a body (e.g. HTTP 204), gives a token-specific hint on HTTP 401/403, and a clear error when the response is not JSON (e.g. an HTML login or captive-portal page).
Added mocked-HTTP tests for af_fetch() and af_fetch_all().
Requires ‘httr2’ >= 1.0.0.

Initial CRAN release.
A generic, dependency-light toolkit for token-authenticated REST APIs, generalising the engine first developed in the ‘BigDataPE’ package.
Token management in process environment variables, namespaced per service: af_store_token(), af_get_token(), af_remove_token(), af_list_tokens().
API profiles via af_api(), with pluggable authentication strategies (af_auth_raw(), af_auth_bearer(), af_auth_header(), af_auth_query()) and pagination strategies (af_paginate_offset(), af_paginate_none()).
Data retrieval with af_fetch() (single page) and af_fetch_all() (chunked, combined into one tibble), built on httr2.
All user-facing output goes through the cli package.
Includes a vignette showing the Big Data PE platform as a worked use case.