fix: replace Venice custom provider with declarative config

[DOsinga]

Summary

Replaces the 566-line custom Venice provider with a ~25-line declarative JSON config.

Problem

The Venice provider had a buggy hand-rolled message formatter that only converted the first tool response per message into a proper role: tool message. When the assistant made multiple tool calls in one turn, the remaining tool results were silently dropped, causing the upstream API to reject the request with:

tool_use ids were found without tool_result blocks immediately after

Solution

Venice is a standard OpenAI-compatible API (/api/v1/chat/completions). Instead of patching the buggy formatter, this PR deletes the entire custom provider and replaces it with a declarative JSON config that delegates to the existing OpenAiProvider. This:

• Fixes the bug — the OpenAI format correctly emits one role: tool message per tool response
• Gains streaming — Venice supports streaming but the custom provider had it hardcoded off
• Removes hardcoded limits — the old code forced temperature: 0.7 and max_tokens: 2048
• Gains all future improvements to the OpenAI format automatically (images, thinking, etc.)

What's dropped (cosmetic only)

• Capability flags appended to model names in the picker (e.g. llama-3.3-70b [cvfsw])
• Custom error message when selecting a model that doesn't support tools
• Configurable VENICE_BASE_PATH and VENICE_MODELS_PATH (nobody uses these)

Changes

• + crates/goose/src/providers/declarative/venice.json — new declarative config
• - crates/goose/src/providers/venice.rs — deleted (566 lines)
• ~ crates/goose/src/providers/mod.rs — removed pub mod venice
• ~ crates/goose/src/providers/init.rs — removed VeniceProvider registration

Fixes #7444