`definedBy` shall be defined separately for RV32/RV64 in instruction yaml (e.g. ld.yaml)

[FanShupei]

Current value and location
Look at a concrete case spec/std/isa/inst/I/ld.yaml:

name: ld
long_name: Load doubleword
description: |
  For RV64, load 64 bits of data into register `xd` from an
  address formed by adding `xs1` to a signed offset.
  <% if ext?(:Zilsd) %>
  For RV32, Loads a 64-bit value into registers xd and xd+1.
  The effective address is obtained by adding
  register xs1 to the sign-extended 12-bit offset.
  <% end %>

definedBy:
  anyOf: [I, Zilsd]

It implies RV32I defines ld instructions. This is absolutely inaccurate.

Expected value

I believe the root cause is current schema forces us to have one "definedBy" for both rv32/rv64. I propose we should extend to schema to allow define different "definedBy" for rv32/64, like the "encoding" field.

Here is an expected "definedBy" for "ld" if the proposal is accepted.

definedBy:
  RV32: Zilsd
  RV64: I

[dhower-qc]

That's a good point. Functionally, what we have is correct because the operation() raises an IllegalInstruction exception in RV32 if Zilsd is not implemented. But what you suggest is more clear.

This will be possible to specify once #891 arrives