Use DeepSeek V4 Flash as the Standard LLM

[RANGROO]

Summary

• replace GLM 4.7 with DeepSeek V4 Flash for the seeded Standard LLM connector
• mark V4 Flash as reasoning-capable so the existing driver explicitly sends reasoning.enabled=false instead of allowing DeepSeek's default high-effort thinking mode
• keep the 750-token, temperature 0.6 JSON-schema configuration; Fast, Powerful, and Experimental remain unchanged

Compatibility

No migration forces an existing user-selected connector to change.

Validation

• PHP lint passed for conf/conf.sample.php, lib/core/llm_connector.class.php, and ui/quickstart.php
• disposable PostgreSQL load confirmed connector ID 1 resolves to DeepSeek V4 Flash with reasoning_model=1, 750 max tokens, JSON/schema enabled, and temperature 0.6
• live CHIM connector probe sent reasoning: { exclude: true, enabled: false }, returned valid action JSON in 5.349 seconds, and OpenRouter reported reasoning_tokens: 0; the prior omitted-reasoning probe took 9.112 seconds
• git diff --check passed
• CHIM unstable push guard: PR REQUIRED for core runtime paths

Deployment

The branch was not deployed. The live probe used a temporary connector matching these defaults in the existing local CHIM runtime; the connector and disposable database state were removed afterward. In-game behavior remains untested.

Related PRs

• Use DeepSeek V4 Flash as the Standard LLM Dwemer-Dynamics/StobeServer#84
• Use DeepSeek V4 Flash as the Standard LLM Dwemer-Dynamics/DialecticServer#65