PR22819, powerpc gas "instruction address is not a multiple of 4"

PR22819, powerpc gas "instruction address is not a multiple of 4"

Checks for insn alignment were hopelessly confused when misaligned
data starts a new frag.  The real-world testcase happened to run out
of frag space in the middle of emitting a trace-back table via
something like:
	.byte 0		/* VERSION=0 */
 	.byte 9		/* LANG=C++ */
	.byte 34	/* Bits on: has_tboff, fp_present */
	.byte 64	/* Bits on: name_present */
	.byte 128	/* Bits on: stores_bc, FP_SAVED=0 */
	.byte 0		/* Bits on: GP_SAVED=0 */
	.byte 2		/* FIXEDPARMS=2 */
	.byte 1		/* FLOATPARMS=0, parmsonstk */
	.long 0
	.long 768	/* tb_offset: 0x300 */
	.hword 45	/* Function name length: 45 */
 	.long 0x334e5a5f
	.long 0x31766f70
	.long 0x65744932
	.long 0x69746172
	.long 0x7a5f6e6f
	.long 0x64504533
	.long 0x5f534e50
	.long 0x72463431
	.long 0x61746361
	.long 0x74535f6c
	.long 0x74637572
	.byte 0x45
	.byte 0
The trigger being those misaligned .long's output for the function
name.  A most horrible way to output a string, especially considering
endian issues..

	PR 22819
	* config/tc-ppc.c (md_assemble): Rewrite insn alignment checking.
	(ppc_frag_check): Likewise.
	* testsuite/gas/ppc/misalign.d,
	* testsuite/gas/ppc/misalign.l,
	* testsuite/gas/ppc/misalign.s: New test.
	* testsuite/gas/ppc/misalign2.d,
	* testsuite/gas/ppc/misalign2.s: New test.
	* testsuite/gas/ppc/ppc.exp: Run them.