I just faced the same problem. The issue is that the when you do 

JobSettings.as_dict()

the settings are parsed to a dict where all the values are also parsed recursively. When you pass the parameters as **params, the create method again tries to parse each parameter to a dict since it needs to create a full json body to make the API request. 
The trick is to parse only the first level of JobSettings to dict (not recursively) and create the params dict.

Try this:

import os
import time

from databricks.sdk import WorkspaceClient
from databricks.sdk.service import jobs
from databricks.sdk.service.jobs import JobSettings

w = WorkspaceClient()

notebook_path = f'/Users/{w.current_user.me().user_name}/sdk-{time.time_ns()}'

cluster_id = w.clusters.ensure_cluster_is_running(
    os.environ["DATABRICKS_CLUSTER_ID"]) and os.environ["DATABRICKS_CLUSTER_ID"]

params = {
    "name":f'sdk-{time.time_ns()}',
    "tasks": [jobs.Task(description="test",
                existing_cluster_id=cluster_id,
                notebook_task=jobs.NotebookTask(notebook_path=notebook_path),
                task_key="test",
                timeout_seconds=0)],
    'email_notifications': jobs.JobEmailNotifications(
        no_alert_for_skipped_runs=True, 
        on_failure= ['some@email.com']
        ),
}
settings = JobSettings(
    **params
)

# This only creates a dictionary but preserving the types of the values.
# Go to https://github.com/databricks/databricks-sdk-py/blob/33b209bdbaca4a956c7375fd65cbf8d6c3b4d95d/databricks/sdk/service/jobs.py and check that some settings has it's own classes (types) with its own method .to_dict()
create_params = {k:getattr(settings,k) for k in settings.as_dict().keys()}


created_job = w.jobs.create(**params) # this works
created_job = w.jobs.create(**settings.as_dict()) # this does not
created_job = w.jobs.create(**create_params) # this does