v2.1.0

feat: default OpenAI to gpt-6-luna (2.1.0) (#5)

* feat: default OpenAI to gpt-6-luna (2.1.0)

OpenAI removes gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini from the API on
14 January 2027 and names gpt-6-luna as the replacement. The bundle's OpenAI
default was gpt-4o-mini, so an application that does not set
providers.openai.default_model would have stopped working on that date without
changing a line of its own code.

Requires ai-sdk ^2.4, which is where the GPT-6 models are defined. That is not
optional: GPT_6_LUNA does not exist on an older SDK, so the constraint is what
keeps this default resolvable rather than fatal.

ai:test-rate-limiting builds its mock client on gpt-6-luna too. It issues no
real request; the id simply stops naming a model that is going away.

README and the rate limiting example stop teaching gpt-4o-mini for the same
reason.

* fix: drop the unused $output PHPStan 2.3 reports

executeMockTest() took an OutputInterface it never used, writing through
SymfonyStyle instead. PHPStan 2.3 flags it and 2.2 did not, and CI resolves
PHPStan fresh because this repo keeps no lockfile, so origin/main was already
failing the same way before this branch existed.