See github issue #62.

Avoid adding unnecessary padding.

Avoid adding unnecessary padding and simplify the code.

Add functions to align a given value to a given alignment and to
calculate the necessary padding to align the value.

Theere were multiple trivial errors in the
handling of buffers, such as putting an extra
"&" to get the address and not copying the
read data back to user.

The smpr_[write/read]_block () functions
return the number of data bytes actually
written. So, we must follow the same
convention on checking for errors.

We must read/write to/from 32-bit registers,
but we can specify how many bytes to read/write
via "charlen" register. So, align size to 32-bit
sizes and handle it according to "charlen".

There are unused address bits, so even if
we can use it, the instruction size is 16-bits
and we must extend it with 0s

We must account for cases when the user
did not "correctly" specified data pointer
and/or data size.

Now, the protocol is "more" independent of the
higher layer. So, any specific changes to it
must be done explicitly by the caller.

Now, we use the offs parameter as the address
to be written by the protocol. This fixes the
protocol semantics, and makes it easier to support
new protocols.

Some chips have 8, 16 or 24-bit (and more )address
size definitions. So, allow the higher layer to
specify its size.

In this way, the only option to get the operations
table for a particular "protocol" is to instantiate
the appropriate class and use the *_get_ops ()
function.

With this, we can be safe that the user will always
pass the pointer ops that is inside the "protocol"
structure (along with its options).

Now, the same information needs to be read from
the appropriate structure through the access methods.

The idea here is to have SPI/I2C-obj objects
to customize the objects dynamically, instead
of having a "flags" parameter that needs knowledge
of the inner protocol.

Now, we pass the pointer to the operation
we want to use it, instead of the enum.

This will make it easier to declare a new operation
and use it. In the previous way, we had to manually
add the operation to the enumeration.

Now, we pass the pointer to the operation
we want to use it, instead of the enum.

This will make it easier to declare a new operation
and use it. In the previous way, we had to manually
add the operation to the enumeration.

As of now, devices that do not support SDB will fail,
as it does not implement a timeout mechanism.

For instance, Ethernet devices will wait forever to
receive the SDB magic number, but this will never happen.
So, avoid creating SDB in the first place until we have
a proper timeout/failover mechanism.

This HALT bit was unused. So, we cahnged its meaning
to "print without prefix/headers", which is useful
for more complex printing, like SDB records.