[core] Split blob-as-descriptor into read-blob-as-descriptor and write-blob-from-descriptor

[Zouxxyy]

Purpose

There are two main reasons for this change:

1. blob-as-descriptor is highly ambiguous during writes—it actually means that the input for writing is a descriptor, not that the blob itself is being written as a descriptor.

2. A single configuration cannot adequately serve both use cases. I believe, the most common scenario should be:

• write-blob-from-descriptor=true so that data is loaded at write time which is extremely memory-efficient. This maybe can be set to default true, in fact, there has been discussion on this topic: [api] Set blob-as-descriptor default to true #7021
• read-blob-as-descriptor=false since in most read scenarios in some case, we need the original raw data rather than a descriptor.

Tests

API and Format

Documentation

[leaves12138]

Thanks for @Zouxxyy , can you add a fallback configuration to make the old table could work the this change?

[Zouxxyy]

Thanks for @Zouxxyy , can you add a fallback configuration to make the old table could work the this change?

I should have added withFallbackKeys("blob-as-descriptor")

[JingsongLi]

We need to conduct more research.