Reduce test time with a global LLM mock

[garylin2099]

Features

• Reduce test time to less than 15 mins
• Almost all tests are mocked, meaning possible to automate run on every PR, which will be very useful to discover problems early
• Responses from passed tests only, which we deem as expected LLM responses, are saved for future reuse.
• Mock avoid providers folder which allows model APIs to be tested differently

Result
https://github.com/garylin2099/MetaGPT/actions/runs/7409829788/job/20160838711

Limitation
Still some failed tests (21 / 375, 6%), need to solve all of them for the tests to be of truly useful

[geekan]

LGTM