test multiple dtypes, and correct the specification of the vector length in the condition_mapper_wrapper

This is achived via:

1.  A new VectorLiteral class, based off ArrayLiteral, but with the string mapping changed to have enclosing parenthesis rather than curly braces.
2.  A new VectorTypeCast class, based off TypeCast, but explicitly storing both the base numpy dtype of the VectorLiteral (for inferencing) and the stringified vector dtype (e.g., 'int4') for generation
3.  Modification of the condition mapper wrapper in `wrap_in_if` to properly generate the VectorTypeCast

and is tested in test/test_loopy.py::test_explicit_simd_vector_iname_in_conditional

Preinstall pybind11 for pyopencl

See merge request !277

Fix legacy PyOpenCL CI

See merge request inducer/loopy!275

Delete dead code surrounding ArrayBase.strides attribute (Closes #144 on gitlab)

Closes #144

See merge request inducer/loopy!274

Fix missing argument to format string

See merge request inducer/loopy!273

Use cgen version in data model version

See merge request inducer/loopy!272

Array arg plus renaming

See merge request !270

1.  Literal to vector assignment (no explicit conversion should be used)
2.  Non-vector temporary variable to vector assignment (no explicit conversion)
3.  Vector-temporary to vector variable assignment (explicit conversion)

Previously, only vector to vector assignment was tested.

This reverts commit 61653be0.

incorporate two different privitizing 'flavors', one for ILP and one for vectorization -- the main difference is that vectorization requires a _direct_ write-dependency on the vectorizing iname, whereas an ILP promotion simply requires that the temporary is inside an ILP-loop