[FEA] Use kvikIO for host_reads and host_writes

[vuule]

We currently only use kvikIO for the device_read and device_write calls, even though kvikIO implements parallel IO to/from host memory as well.

For reads:
By default, we perform memory mapping of the input file, and this is something that kvikIO does not support ATM. We should use kvikIO for host_reads when memory mapping is disabled, to benefit from multi-threaded reads.
We can also evaluate defaulting to kvikIO instead to memory mapping.

For writes:
We should default to kvikIO use, there are no internal optimizations that might be faster than kvikIO.

• host_read
• host_write
• Evaluate memory mapping as the default.