fix(llms): classify gpt-5 dash family as reasoning models

[RahilOp]

Problem

The default OSS config resolves model=None to gpt-5-mini, which is a reasoning model that rejects any non-default temperature. The existing _is_reasoning_model heuristic only matched the exact string gpt-5, so gpt-5-mini fell through to the temperature path. This caused every default Memory().add(...) to fail LLM extraction and silently store nothing.

Closes #6085

Changes

• Updated mem0/llms/base.py to classify the GPT-5 dash family (gpt-5-mini, gpt-5-nano, and dated variants) as reasoning models.
• Explicitly excluded gpt-5-chat* variants, which are non-reasoning chat models that still accept a custom temperature.
• Added regression tests covering the classification and the end-to-end API call (no temperature is sent for the default gpt-5-mini).

Testing

MEM0_TELEMETRY=False hatch run pytest tests/llms/test_openai.py::test_is_reasoning_model_classification tests/llms/test_openai.py::test_default_gpt5_mini_omits_temperature -x

All tests pass. ruff check also passes.

Checklist

• I have read the CLA Document and I sign the CLA

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

Hi @RahilOp, thanks for opening this pull request.

This is just a soft check: you are not yet in this repo's vouched contributor list (.github/VOUCHED.td). Nothing is blocked and there is nothing you need to do.

A maintainer can vouch for you by commenting !vouch @RahilOp on any issue.

[dlowzzxx]

Verified mem0/llms/base.py:75-87 correctly fixes #6085 where default config resolves model=None to gpt-5-mini — a reasoning model that rejects non-default temperature — so the old exact in reasoning_models check caused Memory().add() to silently fail and store nothing. The added if base_model.startswith("gpt-5-") and not base_model.startswith("gpt-5-chat"): return True cleanly partitions the dash family (gpt-5-mini, gpt-5-nano, dated variants with rsplit provider stripping) from non-reasoning gpt-5-chat* and dot-versioned gpt-5.4* which must keep temperature per tests/llms/test_openai.py:290. Reproduced locally that main returns False for gpt-5-mini and confirmed new tests tests/llms/test_openai.py:323-355 cover both paths. Narrow dash-prefix with explicit chat exclusion is the right minimal heuristic — LGTM