Adding GPU support documentation

[NewtonSander]

Introduction

This PR documents how a user can run the NDK simulation using the GPU #32 .
This code was tested in an AWS instance with NVIDIA HPC preinstalled:

PLATFORM='nvidia-acc' python plot_full_scenario.py 
creating a grid with shape: (241, 141) for extent: [0.12 0.07] m
Estimated time to complete simulation: 47 seconds. Memory required is 8 GB (available 64.264081408 GB). These values are approximated.
Default Devito configuration:
	 * autotuning=off
	 * develop-mode=False
	 * mpi=False
	 * log-level=DEBUG
	 * language=openmp
(ShotID 0) Preparing to run state for shot
(ShotID 0) Estimated bandwidth for the propagated wavelet 0.497-0.502 MHz
(ShotID 0) Spatial grid spacing (0.500 mm | 5.981 PPW) is below dispersion limit (0.598 mm | 5.000 PPW)
(ShotID 0) Time grid spacing (0.083 μs | 46%) is above OT2 limit (0.080 μs) and below OT4 limit (0.145 μs)
(ShotID 0) Selected undersampling level 4
(ShotID 0) Selected time stepping scheme OT4
Operator `acoustic_iso_state` instance configuration:
	 * subs={h_x: 0.0005, h_y: 0.0005}
	 * opt=advanced
	 * compiler=pgcc
	 * language=openacc
	 * platform=nvidiaX

Changes

A PLATFORM environment variable is read when running the pde, by setting it to nvidia-acc the GPU can be used to run the simulation