[Backtracing][Linux] Handle inaccessible memory properly in Docker.

[al45tair]

Thanks to a missing - sign, we were returning garbage rather than indicating that the memory region in question was inaccessible. This mainly affects register dumps (since that's the time we expect to have to cope with out-of-bounds reads).

rdar://117900760

[al45tair]

Explanation: The memserver_read() function is supposed to return negative numbers on failure. For normal execution on Linux, it uses process_vm_readv when possible because that is faster and doesn't require using signal handlers to trap failed reads, but when we can't do that (for instance in Docker), it falls back on a memcpy() based alternative. Unfortunately it looks like I fat-fingered the return code when it fails and returned 1 instead of -1, with the result that the memory server would return garbage instead of indicating that the read failed.
Risk: Low. Only really affects register dump output since that's the only time we're really relying on being able to attempt reads to random addresses.
Original PR: #69635
Reviewed by: @tomerd
Resolves: rdar://117900760
Tests: There isn't really a test in the repo for this specific issue.

[al45tair]

@swift-ci Please test

[al45tair]

@swift-ci Please test Linux platform