resolves #4478 (BA-1416).
This PR introduces new parameter allow_fractional_resource_fragmentation, on session creation API's resource_opts input. If the value is not specified on creation API request, it will fall back to to default value as specified on the target scaling group's scheduler_opts.allow_fractional_resource_fragmentation value, which eventually defaults as true.
Setting this as the (ultimate) default true will continue to work as before - allowing fragmentation of AI resources defined as fractional (e.g. cuda.shares). Specifying this as false will block the fragmentation with FractionalResourceFragmented error (see screenshot below).

Testing this PR requires a Backend.AI cluster with two or more accelerators with fractional scaling enabled. If you do not have access to such hardware, using mock accelerator will be also sufficient.

Checklist: (if applicable)

• Mention to the original issue

📚 Documentation preview 📚: https://sorna--4533.org.readthedocs.build/en/4533/

📚 Documentation preview 📚: https://sorna-ko--4533.org.readthedocs.build/ko/4533/