Add :sse_middleware option to allow custom SSE middleware for streaming

[nvelasco]

Problem

Tesla.Middleware.SSE assumes each HTTP chunk is a complete, parseable SSE message. Some LLM providers occasionally send partial chunks (e.g. a bare {) as the first HTTP frame of a streaming response, causing it to crash with:

  %FunctionClauseError{module: Tesla.Middleware.SSE, function: :decode_body, arity: 1}

Change

Introduces a :sse_middleware option on LlmComposer.Settings that allows consumers to inject a custom SSE middleware instead of the default Tesla.Middleware.SSE. The value is forwarded through ProvidersRunner to HttpClient where the middleware stack is assembled.

Default behaviour is preserved — if :sse_middleware is not set, Tesla.Middleware.SSE is used as before.

Usage

  settings = %LlmComposer.Settings{
    stream_response: true,
    sse_middleware: [{MyApp.CustomSSEMiddleware, only: :data}],
    providers: [{LlmComposer.Providers.Google, [model: "gemini-flash"]}]
  }

[nvelasco]

I'm going to recreate the PR
Don't know why but the CI doesn't end

[sonic182]

if something doesn't work, just restart (close and reopen pr) 😄