[TESTS] Use FP32 inference precision, FP16 KV cache precision for pipelines

[ilya-lavrenov]

OpenVINO plugins enable different kind of optimizations by default like KV cache compression to int8, fp16 inference precision, while in GenAI tests we want to test pipelines and how they are compared against HF / optimum w/o extra optimizations:

openvino.genai/tests/python_tests/common.py

Lines 318 to 325 in 4db67ae

	def get_default_properties():
	import openvino.properties.hint as hints
	import openvino as ov
	return {
	hints.inference_precision : ov.Type.f32,
	hints.kv_cache_precision : ov.Type.f16,
	}

Hopefully, we can merge int8 KV cache by default for CB then #1206, because in tests we will still compare FP16 KV cache, while official Validation should be responsible for validation against reference via WWB metrics.