fix off-by-one write in loader_platform_executable_path

[aizu-m]

ASAN on a minimal harness of the Linux readlink branch (buffer sized to the caller's 1024):

WRITE of size 1 at 0x619000000980 thread T0
    #0 loader_platform_executable_path
0x619000000980 is located 0 bytes after 1024-byte region [...980)

readlink() never appends a terminator and returns up to size bytes. When /proc/self/exe resolves to a path of 1024 bytes or more it returns exactly 1024, and buffer[count] = '\0' then writes one byte past current_process_path[1024], the stack array get_loader_settings hands in.

Found this while diffing the exe-path helpers against each other. The QNX branch already bails on rdsize == size, and the comment above this function promises a NULL return for over-long paths, but the readlink branch never enforced either. Bail out when count reaches size so the terminator always has room, which also makes the code match its documented behaviour.

[ci-tester-lunarg]

Author aizu-m not on autobuild list. Waiting for curator authorization before starting CI build.

[ci-tester-lunarg]

Author aizu-m not on autobuild list. Waiting for curator authorization before starting CI build.

[ci-tester-lunarg]

CI Vulkan-Loader build queued with queue ID 24487.

[ci-tester-lunarg]

CI Vulkan-Loader build # 3602 running.

[ci-tester-lunarg]

CI Vulkan-Loader build # 3602 passed.