feat(client): forward max_retries through Client/ClientV2 constructors

[fern-support]

Summary

Makes max_retries configurable from the client constructors. Requested by Cohere (handle retries centrally outside the SDK; the default of 2 made that hard to enforce).

The base BaseCohere / AsyncBaseCohere constructors already accept max_retries, but the hand-maintained override constructors in client.py / client_v2.py never forwarded it — so max_retries could only be set per-request via request_options, not at the client level. (These files are .fernignore'd, so fern generate never touches them; this had to be a manual edit.)

Changes

Thread max_retries: typing.Optional[int] = None through all four override constructors, forwarding to the respective super().__init__:

• Client → BaseCohere
• AsyncClient → AsyncBaseCohere
• ClientV2 → Client
• AsyncClientV2 → AsyncClient

The None default preserves existing behavior: the base client applies its own default of 2. Per-request max_retries in request_options continues to take precedence.

Usage

import cohere

client = cohere.AsyncClientV2(api_key="...", max_retries=0)  # disable SDK-level retries

Verification

• inspect.signature(...).parameters contains max_retries for Client, AsyncClient, ClientV2, AsyncClientV2.
• Instantiating each with max_retries=0 stores 0 on the client wrapper; instantiating without it stores the base default of 2.
• Confirmed timeout is still forwarded for all four (regression guard).
• mypy src/cohere/client.py src/cohere/client_v2.py — clean.
• Offline unit tests pass (test_client_init, test_overrides, test_embed_utils, test_embed_streaming).

---

Note

Low Risk
Constructor-only plumbing with None preserving prior defaults; no change to retry logic itself.

Overview
Adds optional max_retries to Client, AsyncClient, ClientV2, and AsyncClientV2 so callers can set SDK-level retry behavior at construction (e.g. max_retries=0 to disable the default of 2).

The hand-maintained overrides in client.py and client_v2.py now accept max_retries: typing.Optional[int] = None and pass it through to BaseCohere / AsyncBaseCohere (and V2 wrappers delegate via Client / AsyncClient). Omitting the argument keeps the existing base default; per-request request_options max_retries still overrides the client default.

^{Reviewed by Cursor Bugbot for commit df151d6. Bugbot is set up for automated code reviews on this repo. Configure here.}