Add internal slots to MLOperand, MLActivation and basic algorithms

[zolkis]

In the validation steps for various operations (for instance Batch Norm), the following dependencies/references/algorithms need to be defined (this entry will be kept updated):

• Internal slots for MLOperand:
  • [[type]] as MLOperandType
  • [[dimensions]] as sequence<unsigned long>
  • [[name]] as string (only for input operands)
  • [[data]] as ArrayBufferView (tensor of given dimensions)
  • [[builder]] the associated MLGraphBuilder
• Algorithm (or internal slot) to get the rank of the operand (for checking range for axis)
• Algorithm to internally create an MLOperand given |builder| and |descriptor|, used by op algorithms.
• Algorithm to internally copy an MLOperand, used by op algorithms.
• Algorithm to check dimensions given |dimensions| and |type|.
• Internal slots for MLActivation:
  • [[name]] as string, e.g. "relu", "sigmoid", "softmax", etc.
  • ? [[input]] as MLOperand (needed?)
  • ? [[output]] as MLOperand (intermediate output?)
• Algorithm to create an MLActivation, used by the op algorithms.

Related to the clarification of terms requested in #335.

[anssiko]

@zolkis given PR #400 (comment) was closed please provide a record here of commit(s) and/or PR(s) that address this issue. When done, close this issue.

My recommendation is to link each PR with an issue using the Fix keyword either in the PR description of the commit message to enable the associated issue(s) to be automatically closed upon merge.

[zolkis]

The content of #400, and previously, #337, has been agreed with @huningxin to be split among the PRs that will have dependencies to this content:
#347, #348, #398 (#366), #365