Simplify task switches

No longer using the rsp from the entry to the kernel, but instead
switching rsp at task-switching time in assembly.

This currently breaks fork()

src/drivers/nulldrv/main.cpp

@@ -3,6 +3,7 @@
 
 extern "C" {
 	int32_t getpid();
+	int32_t fork();
 	void sleep(uint64_t til);
 	void debug();
 
@@ -14,6 +15,7 @@ int
 main(int argc, const char **argv)
 {
 	int32_t pid = getpid();
+	//int32_t child = fork();
 	debug();
 	for (int i = 1; i < 5; ++i)
 		sleep(i*10);

src/drivers/nulldrv/main.s

@@ -39,6 +39,18 @@ sleep:
 	pop rbp
 	ret
 
+global fork
+fork:
+	push rbp
+	mov rbp, rsp
+
+	mov rax, 8
+	syscall    ; pid left in rax
+
+	pop rbp
+	ret
+	
+
 global _start
 _start:
 	xor rbp, rbp	; Sentinel rbp