Populate reasoning_levels and default_reasoning_effort on the
Anthropic-on-Bedrock entries for the three models that support
the effort parameter, and correct the direct-Anthropic entries
to match Anthropic's own documentation.

Per https://docs.anthropic.com/en/docs/build-with-claude/effort:

  - sonnet-4-6:  levels=[low,medium,high,max],          default=high
  - opus-4-7:    levels=[low,medium,high,xhigh,max],    default=high
  - opus-4-6:    levels=[low,medium,high,max],          default=high

Changes vs. the previous state:

  - bedrock/anthropic.claude-sonnet-4-6: added levels + default
  - bedrock/anthropic.claude-opus-4-7:   added levels + default (incl xhigh, which is Opus-4.7-only per docs)
  - bedrock/anthropic.claude-opus-4-6-v1: added levels + default
  - anthropic/claude-opus-4-7:           added xhigh to levels (Opus 4.7 is the only model that exposes xhigh)
  - anthropic/claude-{sonnet-4-6,opus-4-7,opus-4-6}: default_reasoning_effort medium -> high

The API default per Anthropic's docs is "high" ("Setting effort to \"high\"
produces exactly the same behavior as omitting the effort parameter entirely"),
so the previous "medium" default did not match server behavior.

Co-authored-by: JJ Bot <john.jansen+bot@me.com>

Devstral-Small-24B is being deprecated from the Neuralwatt API.
Qwen3.6-35B is the designated replacement for the small model tier.
Only touches neuralwatt provider files.

The previous commit added K2.6 as a model but didn't update the
default_large_model_id or the feature matcher, causing
TestValidDefaultModels to fail.

💘 Generated with Crush

Assisted-by: GLM-5.1 via Crush <crush@charm.land>

This reverts commit 2b450ae9a4148a38b2a3c5aec6b783ae0227f018.

- Update API endpoint to https://api.deepseek.com (remove /v1 suffix)
- Add deepseek-v4-flash model (supports both thinking/non-thinking modes)
- Add deepseek-v4-pro model with full feature support
- Update context window to 1M tokens for all models
- Add new capabilities: json_output, tool_calls, conversation_prefix, fim
- Mark deepseek-chat and deepseek-reasoner as deprecated (2026-07-24)
- Update pricing to current official rates

💘 Generated with Crush

Assisted-by: MiniMax-M2.7 via Crush <crush@charm.land>

And add some new models as a consequence