chmod should be a read-modify-write of global meta data

Nobody has claimed this yet.

The chmod wrapper updates the global file meta to record new permissions of the file. In case some other process has changed the global meta data, this probably should be a read-modify-write of the global meta data. The current process may have stale info, e.g.

```
// if the file has "rw-" bits set
if (rank == 0) {
MPI_Barrier(COMM_WORLD)
chmod(path, "-r")
} else if (rank == 1) {
chmod(path, "+x")
MPI_Barrier(COMM_WORLD)
}
// here it should now be "-wx"
// rank 1 adds "x", then rank 0 drops "r"
```
Without the read-modify-write, we'd get "-w-" because rank 0 wouldn't see the change by rank 1.

Start by locating the C chmod wrapper and the code that updates global file metadata, then trace how metadata is read and written around concurrent chmod calls. Reproduce the two-rank example if the project provides an MPI test setup; done means each chmod preserves changes made by another process, producing the combined permissions rather than overwriting them.

Written by the indexing model from the issue text.