Stop APE bare metal loader from reading beyond program image end

This allows e.g. `qemu-system-x86_64 -s o/examples/hello.com
-serial stdio` to work without having to add extra padding to
the end of the `hello.com` "disk image".

(The sector count computation is divided among two instructions
in the assembly code.  This is done on purpose, to prevent an
ASCII 0x27 (single quote) byte from appearing in the bare
metal loader code, which will break the shell script loader.
There is probably a better way.)

ape/ape.S

@@ -250,6 +250,7 @@ pc:	cld
 	xor	%cx,%cx				# current cylinder
 	xor	%dh,%dh				# current head
 	mov	$v_ape_realsectors,%di		# total sectors
+	sub	$v_ape_realslacksectors,%di
 3:	call	pcread
 	dec	%di
 	jnz	3b
@@ -1567,6 +1568,7 @@ kernel:	movabs	$ape_stack_vaddr,%rsp
 	.ldsvar	_end
 	.ldsvar	_etext
 	.ldsvar	v_ape_realsectors
+	.ldsvar	v_ape_realslacksectors
 	.ldsvar	v_ape_highsectors
 	.ldsvar	ape_idata_ro
 	.ldsvar	ape_pad_rodata

ape/ape.lds

@@ -561,6 +561,9 @@ SHSTUB2(ape_loader_dd_count,
 HIDDEN(v_ape_realsectors =
            MIN(0x70000 - IMAGE_BASE_REAL,
                ROUNDUP(RVA(_edata), 4096)) / 512);
+HIDDEN(v_ape_realslacksectors =
+           v_ape_realsectors - MIN(0x70000 - IMAGE_BASE_REAL,
+                                   ROUNDUP(RVA(_edata), 512)) / 512);
 HIDDEN(v_ape_realpages = v_ape_realsectors / (4096 / 512));
 HIDDEN(v_ape_highsectors =
            (ROUNDUP(RVA(_edata), 512) / 512) - v_ape_realsectors);