DSH RLM 0.1.2 — responsive provider setup

DSH RLM 0.1.2 — responsive provider setup

Experimental macOS (Apple Silicon/Intel) and glibc Linux (ARM64/x86-64). All four platform builds and relocated wheel smoke tests passed GitHub Actions run 36262445755. Matching platform bundles are attached; the same uvx command automatically selects the correct one.

Provider directory now renders independently of settings, credentials, and model discovery. Models load on provider selection through an authenticated provider-scoped endpoint; setup no longer waits for every provider catalog. Loading states stay explicit and authentication revalidation remains fail-closed.

Chrome tests with five-second catalog/auth delays showed providers in about 0.6 seconds after page navigation. These timings are local test evidence, not a startup performance guarantee. Stock DSH UI still makes separate global catalog requests. No actual provider login/inference tested.

Stop the old process, then uvx --upgrade dsh-rlm@0.1.2 --port 3081 after matching PyPI publication. No cache deletion needed.