pmemcheck incosistently reporting issues w.r.t. stores outside of transactions

[pbalcer]

This code:

	VALGRIND_REMOVE_FROM_TX(run, runsize);

	pmemops_persist(&heap->p_ops, run->bitmap, sizeof(run->bitmap));

	struct chunk_header run_data_hdr;
	run_data_hdr.type = CHUNK_TYPE_RUN_DATA;
	run_data_hdr.flags = 0;

	struct chunk_header *data_hdr;
	for (unsigned i = 1; i < m->size_idx; ++i) {
		data_hdr = &z->chunk_headers[m->chunk_id + i];
		VALGRIND_DO_MAKE_MEM_UNDEFINED(data_hdr, sizeof(*data_hdr));
		VALGRIND_ADD_TO_TX(data_hdr, sizeof(*data_hdr));
		run_data_hdr.size_idx = i;
		*data_hdr = run_data_hdr;
		VALGRIND_REMOVE_FROM_TX(data_hdr, sizeof(*data_hdr));
	}
	pmemops_persist(&heap->p_ops,
		&z->chunk_headers[m->chunk_id + 1],
		sizeof(struct chunk_header) * (m->size_idx - 1));

	struct chunk_header *hdr = &z->chunk_headers[m->chunk_id];
	ASSERT(hdr->type == CHUNK_TYPE_FREE);

	VALGRIND_ADD_TO_TX(hdr, sizeof(*hdr));
	struct chunk_header run_hdr;
	run_hdr.size_idx = hdr->size_idx;
	run_hdr.type = CHUNK_TYPE_RUN;
	run_hdr.flags = header_type_to_flag[c->header_type];
	*hdr = run_hdr;
	VALGRIND_REMOVE_FROM_TX(hdr, sizeof(*hdr));

*data_hdr = run_data_hdr; is the problematic assignment.

Produces the following issue:

==28070== Number of stores made without adding to transaction: 14
==28070== Stores made without adding to transactions:
==28070== [0]    at 0x4C471F7: heap_run_init (heap.c:393)
==28070==    by 0x4C47AFE: heap_create_run (heap.c:517)
==28070==    by 0x4C48A01: heap_ensure_run_bucket_filled (heap.c:825)
==28070==    by 0x4C48DF0: heap_get_bestfit_block (heap.c:909)
==28070==    by 0x4C5AD50: palloc_operation (palloc.c:211)
==28070==    by 0x4C5C1D0: pmalloc_operation (pmalloc.c:116)
==28070==    by 0x4C5C3F7: pmalloc_construct (pmalloc.c:166)
==28070==    by 0x4C648E1: tx_alloc_common (tx.c:1010)
==28070==    by 0x4C679B1: pmemobj_tx_alloc (tx.c:1961)
==28070==    by 0x404F2B: sc8_verify_abort (obj_convert.c:467)
==28070==    by 0x405A67: main (obj_convert.c:594)
==28070== 	Address: 0x6102490	size: 16

But only sometimes (can't reproduce this reliably outside of this specific case: pmem/pmdk#2242).
Regardless of anything else, the report is false, because the store is clearly added to the transaction.

Wrapping the entire code section in a add_to_tx, that encompasses the entire modifiable area, fixes the problem.