Reform chatbot client API code

[pseudotensor]

but serialize=False doesn't work due to gr.state issue: gradio-app/gradio#4081

Original here that had to use json files due to serialize=True default:

h2ogpt/client_test.py

Lines 81 to 107 in 05c742a

	else:
	api_name = '/instruction'
	import json
	foofile = '/tmp/foo.json'
	with open(foofile, 'wt') as f:
	json.dump([['', None]], f)
	args += [foofile]
	if not stream_output:
	for res in client.predict(
	*tuple(args),
	api_name=api_name,
	):
	print(res)
	res_file = client.predict(*tuple(args), api_name='/instruction_bot')
	res = json.load(open(res_file, "rt"))[-1][-1]
	print(md_to_text(res))
	else:
	print("streaming instruction_bot", flush=True)
	job = client.submit(*tuple(args), api_name='/instruction_bot')
	while not job.done():
	outputs_list = job.communicator.job.outputs
	if outputs_list:
	res_file = job.communicator.job.outputs[-1]
	res = json.load(open(res_file, "rt"))[-1][-1]
	print(md_to_text(res))
	time.sleep(0.1)
	print(job.outputs())

[pseudotensor]

==================================================================================== 4 passed, 1 skipped in 310.72s (0:05:10) =====================================================================================
(h2ollm) jon@pseudotensor:~/h2o-llm$