[DAPHNE-#358]: DaphneDSL: Matrix literals

[akroviakov]

Implements Matrix literal for DaphneDSL.

• A light bool type support for DenseMatrix is added (for print and reshape kernels).
• Creates a DenseMatrix with shape (nx1) of type int64_t, double or bool inside DaphneDSLVisitor.
• Usage example: x = [1, 2, 3] or x = [1.1, 2.2, 3.3] or x = [true, false, true].
• Throws error in case of:
• • You try to create a matrix with mixed types (e.g., x = [1, 2.2, 3]).
• • You try to create an empty matrix: can't infer the type.
• • You use a not yet defined in ValueTypeUtils type (e.g., const char*).

Closes #358 .

[pdamme]

Thanks a lot, @akroviakov, it looks very good to me. I've applied just a few minor changes for polishing. Matrix literals are a very useful feature. Having a basic support for them will simplify writing DaphneDSL scripts a lot :) .

I can imagine several follow-up issues relaxing some current constraints, e.g.:

• Empty matrix literals could just assume double as the value type. The user can always cast that to any other value type like as.si64([]) (which will be zero-cost).
• Allowing mixed element types by infering the common value type of the matrix via type promotion (see Type promotion in DaphneDSL #364).
• Allowing complex expressions for the elements of a matrix literal. All elements might become known at compile-time through constant folding, then we could create a MatrixConstantOp in a compiler pass. However, when not all elements are known at compile-time, we could create DaphneIR ops to calculate them at runtime (will look like a matrix literal in DaphneDSL, but be executed differently).

BTW: The initial support for bool as a value type would not have been strictly required, but it doesn't hurt either. I'll leave a comment in the respective issues (#39, #40) ;) .