[DONT MERGE] Fp4 experiment

[cloudhan]

@zjing14 This is a fully correct pipeline that support packed fp4 (2 int4s in a byte). This is used for demonstrate what might need to be changed to support subtype opearation. The conversion however is a dummy opeartion with adhoc impl.

The problem with current ck:

1. no subbyte vector support.
2. no subbyte addressing support.
3. no hierarchical layout description.

[zjing14]

@cloudhan Thx very much for working on sub-byte data type and pointing out issues in CK. We are working on it. @geyyer