Refactor how clients are parsed for more accurate and consistent erro…

…r handling (#1026)

Prior we had a lot of duplicated code for parsing args which caused some
confusion. Adding new error handing logic in.

Included some new tests which ensure prompt-fiddle and baml-cli init
sample code always passes without any warnings!
Python library static analysis fixes. (0 errors now!)

docs/old/docs/snippets/clients/providers/groq.mdx

@@ -16,6 +16,7 @@ client<llm> MyClient {
     base_url "https://api.groq.com/openai/v1"
     api_key env.GROQ_API_KEY
     model "llama3-70b-8192"
+    default_role "user"
   }
 }
 ```

docs/old/docs/snippets/clients/providers/openai-generic.mdx

@@ -63,6 +63,8 @@ client<llm> MyClient {
 
 </ParamField>
 
+<Markdown src="../../../../snippets/allowed-role-metadata-basic.mdx" />
+
 ## Forwarded options
 
 <ParamField

engine/baml-lib/baml/build.rs

@@ -41,6 +41,7 @@ fn build_folder_tests(dir: &'static str, out_file_name: &str) {
     println!("cargo:rerun-if-changed={dir}");
     let mut all_schemas = Vec::new();
     find_all_schemas("", &mut all_schemas, dir);
+    all_schemas.sort();
 
     // concatenate all the files in the directory into a single file
     let mut out_file = fs::File::create(format!("{CARGO_MANIFEST_DIR}/{out_file_name}")).unwrap();