Ray-on-aml

Fixing problem with aiohttp package from raydp.
Removed raydp,xgboost-ray, rayserve from default installation.

**Full Changelog**: https://github.com/microsoft/ray-on-aml/compare/v0.1.9...v0.2

Remove GPU image from Spark and fix ray path in job mode

Fixing problem with 0.1.7 where workers resources are not loaded

In some new CI, the ray command is not recognized with subprocess so a full path to ray is required.
This release also adds support for ray 1.12.0

Fixing issue 12