Disallow using the PC as a base in LDR and instead provide a dedicated assembler
method for pc-relative loads. The reason for this is that the generic
`Assembler::ldr` method may decide to generate more instructions if the offset
is out of range, and if the PC was the base, we would get surprising
results. For example:

~~~
ldr r0, [pc, #0xcabba9e]
~~~

is not equivalent to:

~~~
movw ip, #0xba9e
movt ip, #0xcab
ldr r0, [pc, ip]
~~~

since the reference to the PC has moved down two instructions!

We could teach the assembler to handle those cases correctly, but pc-relative
loads are used in specific cases only so that's not necessary.

As a drive-by, remove a reference to code aging.

Bug: 
Change-Id: I586d83a418db52cf28d3b524f889bf40f077998a
Reviewed-on: https://chromium-review.googlesource.com/847008Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Pierre Langlois <pierre.langlois@arm.com>
Cr-Commit-Position: refs/heads/master@{#50475}