[DT][CPU]: RVV scalable encoding materialization (#24600)

This PR introduces scalable data-tiled layout selection and
materialization for RVV. Though since there exist microkernels for the
RV64 backend with the V extension, previous mechanism to select these
data-tiled layouts _statically_ based on the `+zvl*b` flags is also
preserved and the scalable data-tiled layout materialization is hidden
behind the flag scalable vectorization flag.

The exactl layout selection (the tile sizes) though is based on the same
principles as the static one - fully utilize all 32 of the V registers.
F16 tile sizes that underutilize the V registers in the static case (see
[this discord
thread](https://discord.com/channels/689900678990135345/1456679894812590326/1461764521403224209))).

**NOTE**: the `+zvl*b` flags DO NOT give us the _exact_ values of VLEN
that the hardware implements but rather the minimum one, meaning that
the previous static mechanism can underutilize the vector registers and
is not vector-length agnostic.

---------

Signed-off-by: Ege Beysel <beyselege@gmail.com>