Quantization Verifiers based on T2x set of Traits

[abhigunj]

• made isCompatibleElementTypeForHloTypeInference stricter to return error for {not Quantize, Quantize}, {per-axis Quantized, per-tensor Quantized} cases
• AddOp VHLO Test failures : addressed test failures because {not Quantize, Quantize} is not allowed
• CorrectedTraits for CholeskyOp and ClampOp to match it with the spec

Note: This PR is based on in review PR #2007

Follow up PR will add/update OP verifiers for OPs which need special handling

[sdasgup3]

Few initial comments. Will add more related to per-op changes

[sdasgup3]

Some of the unhandled cases:

• compare op: Per the spec it uses baseline_element_type(lhs) = baseline_element_type(rhs). However, this check is implemented as

  stablehlo/stablehlo/dialect/StablehloOps.td

  Line 1534 in 70db15a

  SameOperandsElementType /*compare_c1*/,

  . How about we add HLO_CompatibleOperandsElementType equivant to
  SameOperandsElementType but work with baseline_element_types.

• triangular_solve: Use HLO_CompatibleOperandsAndResultElementType instead
  of SameOperandsAndResultElementType.

• batch_norm_grad, batch_norm_training, and batch_norm_inference: The
  verifier uses AllElementTypesMatch whereas the spec wants the operand to
  matgch using baseline type. Can we introduce HLO_AllElementTypesCompatible.

• collective_broadcast, collective_permute: They are currently using
  HLO_CompatibleOperandsAndResultType, but per the spec the type checks should
  be implemented using SameOperandsAndResultType

• reshape: Currently uses HLO_CompatibleOperandsAndResultType but per the
  spec needs stricter type checks using SameOperandsAndResultType.

[abhigunj]

Some of the unhandled cases:

• compare op: Per the spec it uses baseline_element_type(lhs) = baseline_element_type(rhs). However, this check is implemented as

  stablehlo/stablehlo/dialect/StablehloOps.td

  Line 1534 in 70db15a

  SameOperandsElementType /*compare_c1*/,

  
  . How about we add HLO_CompatibleOperandsElementType equivant to
  SameOperandsElementType but work with baseline_element_types.
• triangular_solve: Use HLO_CompatibleOperandsAndResultElementType instead
  of SameOperandsAndResultElementType.
• batch_norm_grad, batch_norm_training, and batch_norm_inference: The
  verifier uses AllElementTypesMatch whereas the spec wants the operand to
  matgch using baseline type. Can we introduce HLO_AllElementTypesCompatible.
• collective_broadcast, collective_permute: They are currently using
  HLO_CompatibleOperandsAndResultType, but per the spec the type checks should
  be implemented using SameOperandsAndResultType
• reshape: Currently uses HLO_CompatibleOperandsAndResultType but per the
  spec needs stricter type checks using SameOperandsAndResultType.

Thanks again for the review.
collective_broadcast collective_permute reshape are being handled in follow up PR since existing Trait can't be directly reused for them. Added Traits for other OPs.

[GleasonK]

A few test recommendations to ensure this is all working as expected.