[bug] --dry-run accepts an invalid llm client base_url

[ting-hong-shieh]

Symptom

On current main, a deployment with base_url = "not a url" passes
switchyard-server --dry-run, prints server OK: test-route, and exits 0. The
same deployment also starts normally; its first model request then fails with
HTTP 502 because the upstream request cannot be constructed.

Reproduction

No API key or live upstream is required.

Create routes.toml:

schema_version = 1

[llm_clients.invalid]
format = "openai_chat"
base_url = "not a url"

[targets.invalid]
id = "upstream-model"
llm_client = "invalid"

[routes.invalid]
id = "test-route"
type = "passthrough"
target = "invalid"

Run the dry check:

cargo run --locked -p switchyard-server -- \
  --config routes.toml --dry-run

It exits 0 and prints:

server OK: test-route

Start the same deployment:

cargo run --locked -p switchyard-server -- \
  --config routes.toml --host 127.0.0.1 --port 43173

Then send a request from another terminal:

curl -i http://127.0.0.1:43173/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"test-route","messages":[{"role":"user","content":"hello"}]}'

The response is:

HTTP/1.1 502 Bad Gateway

{"error":{"code":"upstream_configuration_error","message":"failed to build upstream request: builder error","type":"upstream_error"}}

Expected vs. actual

• Expected: --dry-run rejects llm_clients.invalid.base_url as an invalid
  absolute HTTP(S) URL and identifies the affected client. This check should not
  require connecting to an upstream provider.
• Actual: The non-empty value passes configuration validation. Its invalid
  syntax is discovered only when the first request is sent.

Environment

• Switchyard commit SHA: b256d936f1d77bf13ec9bec399ea0a253e07ca05
• Python version: N/A (Rust server path)
• Rust version: rustc 1.96.1 (31fca3adb 2026-06-26)
• OS / arch: macOS 26.5.2 / arm64
• Install path: source build with cargo run --locked
• Inbound format: OpenAI Chat Completions
• Backend: none; the malformed URL prevents an upstream request from being built

Additional context

build_backend
currently checks only whether base_url is empty. The URL reaches
reqwest::Client::post
when a model call is made.

A local syntax check during deployment construction would catch the error while
preserving --dry-run's no-network behavior. Validating the resolved endpoint
with the same URL-joining rules used at request time would also avoid creating a
second interpretation of base_url.

I searched open and closed issues and pull requests for base_url, dry-run,
invalid URL, builder error, and upstream_configuration_error, and found no
matching report.

I'm happy to send a PR for this if the expected validation behavior sounds right.