The 'make export' target needs to bundle up the user C startup file (crt0), not the kernel head object

Space at the beginning of the process data space is now reserved for user heap management structures.  In the kernel build mode, these heap structures are shared between the kernel and use code in order to allocate user-specific data.

There used to be two ways to pass parameters to new tasks, depending upon the configuration:  Either (1) argv[] as created as an array with each string strdup'ed.  Or (1) argv[] array and strings were created on the stack before the new task was started.   Now, there is only one way, way (1).  Way (2) might be slightly more compact, but this is not worth carry the complexity of two different ways of doing the same thing.

Obsolete support for the 8051 family.  Trying to maintain compatibility with this family is too much effort and there has never really been a successful NuttX impplementation on these parts

Clean up some kernel build heap allocation issues.  The Cortex-A kernel build now compiles without errors (but cannot link until brk() and sbrk() are implemented).

Various changes/fixes to get configs/stm32f4discovery/kostest working after the big configuration renaming (and after a long period of bit rot)

Fix hard coded values in dispatch_sysall inline assembly. Back out/corect part of last change; that was going the wrong direction.

Fix ARM7/9 and Cortex-A SYSCALLs:  For threads in SVC mode, the SVC instructions clobbers R14.  This must be taken account in the inline assembly

Rename  arch/arm/src/armv7-a/syscall.h to svcall.h to work around some include path name collisions; fix some compilation errors in SYSCALL logic when debug is enabled

STM32 FLASH fixes: use size_t instead of uint16_t, make interface more generic.  From Freddie Chopin