docs(server): drop the unimplemented lora_adapter request field

[jamesburton]

Option 1 from the issue: remove the stale field and say what to use instead.

Confirmed unimplemented before changing anything — grep -ri "lora_adapter\|LoraAdapter" src/DotLLM.Server/ returns nothing, and in fact grep -i lora src/DotLLM.Server/ returns nothing
at all.

One correction to the issue's premise

The issue says "there is a registry and admin endpoints". There are not. Grepping the whole tree,
LoRA exists only as IAdapterManager (an interface with no implementation) and
InferenceRequest.AdapterId in DotLLM.Engine. No adapter loader, no server endpoint, nothing
mapped. It is roadmap step 47, unstarted. So rather than point at admin endpoints that do not exist,
the doc now points at the honest workaround: merge the adapter into the weights offline and load the
merged model, via --model at startup or POST /v1/models/load — both of which I verified exist
(ServerOptions.cs:89, ModelManagementEndpoint.cs:29).

Changes

• docs/SERVER.md — drops lora_adapter from the chat-completions request example; adds a short
  note on why there is no such field, what the actual state of LoRA support is, and what to do
  instead.
• docs/LORA.md — adds a one-line status header marking it a design spec for roadmap step 47.
  Without this the two docs contradict each other: LORA.md's "Runtime Application" section says
  "Each request specifies lora_adapter", which is the same claim SERVER.md just retracted. It
  reads fine as intended design once labelled as such.

Docs only, no tests. grep -ri lora docs/ shows the remaining hits are ROADMAP/ARCHITECTURE/AOT
references to the planned feature, which stay accurate.

Closes #425

[Copilot]

Pull request overview

This PR updates the server documentation to stop advertising a per-request lora_adapter parameter that is not implemented in DotLLM.Server, and clarifies the current (unimplemented) status of LoRA serving while pointing users to a workable alternative (offline-merged weights + model load).

Changes:

• Remove the stale lora_adapter field from the POST /v1/chat/completions request example in docs/SERVER.md.
• Add an explicit note in docs/SERVER.md explaining that per-request LoRA adapter selection is not supported yet and recommending the offline-merge + --model / POST /v1/models/load workflow.
• Mark docs/LORA.md as a design spec (roadmap step 47), so its “per-request adapter selection” text is clearly described as intended future behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
docs/SERVER.md	Removes unimplemented lora_adapter from request example and documents the current LoRA status + recommended workaround.
docs/LORA.md	Adds a status header clarifying the document is a design spec and not currently implemented.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.