allocman: fix use of ALLOCMAN_NO_PADDR

Throughout allocman, various parts of the code for allocating a
specific paddr would check that the paddr fell within the bounds of a
given untyped node. Unfortunately, in some cases ALLOCMAN_NO_PADDR +
node->size would overlap with the paddr and cause a false positive -
allocman would *think* that a given node contains the request paddr,
when in fact it didn't.

Eventually, everything would fall apart: ALLOCMAN_NO_PADDR+node->size
would not be large enough to reach the paddr, and allocman would end
up with a NULL node where it wasn't expecting one.

Fix this issue by avoiding nodes with paddr == ALLOCMAN_NO_PADDR
whenever we're searching through node lists, thus preventing
such 'false positive' situations from occurring.

trivial: style allocman utspace code

Fix all the whitespace

- remove trailing whitespace
- remove duplicate blank lines
- remove blank lines at end of files

Fix licenses

libsel4allocman: alloc from both untyped pools

When passing canBeDev and paddr == ALLOCMAN_NO_PADDR to
_utspace_split_alloc it should try and allocate a frame from the device
untyped pool if possible, otherwise fall back to regular untypeds.

sel4allocman: emit errors when retype fails

libsel4allocman: Return allocated objects to the correct pool

The split untyped allocator maintains three pools for different kinds of untyped
objects. Previously the origin of an object was not tracked and when the object
was freed (i.e. returned to the allocator) it was just put back into kernel only
untyped pool. This naturally corrupted the book keeping resulting in future
allocations failing. This change records which pool any allocation is sourced
from and returns back to that pool

Remove assumptions about the 0 paddr being invalid

There is no way to prevent platforms from having interesting memory
or devices at paddr 0, as such we should not be using 0 to signal
the absence of a physical address. This change fixes some cases
where we were doing this, and switches to sentinal value that cannot
collide with a valid page base physical address.

libsel4allocman: Fix 32/64bit compatiblities in LOG functions

libsel4allocman: Memory mapped device IO allocation support

Adds support at the top level allocman interface for allocating
objects at a particular physical address (allocman_utspace_alloc_at)
as well as fullfilling allocations from 'device memory'. Uses this
for implementing the additional vka interface function vka_utspace_alloc_at

The split and twinkle untyped allocators are updated to support this
new interface. split implements the 'alloc_at' by brute force searches
over its untypeds, whilst twinkle with its lack of book keeping
is forced to fail any 'alloc_at' request

Remove all references to KERNEL_STABLE

libsel4allocman: More error information

libsel4allocman: Make allocman 64-bit friendly

When compiling for 32-bit there may be spurious warnings due to
some interface type changes. These will eventually be fixed when
other libraries are made 64-bit compatible