nixosTests.installer: Fix grub1 test being unreliable

The kernel sometimes assigns `/dev/sdb` to the 8GiB disk. This, in turn,
means the test will fail because we're targeting the wrong disk.

```
machine # [    0.000000] sd 2:0:0:0: [sda] 16777216 512-byte logical blocks: (8.59 GB/8.00 GiB)
machine # [    0.000000] sd 3:0:0:0: [sdb] 1048576 512-byte logical blocks: (537 MB/512 MiB)
```

```
machine # [    0.000000] sd 2:0:0:0: [sdb] 16777216 512-byte logical blocks: (8.59 GB/8.00 GiB)
machine # [    0.000000] sd 3:0:0:0: [sda] 1048576 512-byte logical blocks: (537 MB/512 MiB)
```

Note how the "sd x:0:0:0:` ID is stable. That is because QEMU **is**
told to give specific identifiers to the disks. So using the
dev/disk/by-id/ identifiers is stable.

* * *

Tested by forcing the sda/sdb swap this way:

    diff --git a/nixos/tests/installer.nix b/nixos/tests/installer.nix
    index 24c55081f9a..2eee224351b 100644
    --- a/nixos/tests/installer.nix
    +++ b/nixos/tests/installer.nix
    @@ -702,12 +702,19 @@ in {
               + " mkpart primary linux-swap 1M 1024M"
               + " mkpart primary ext2 1024M -1s",
               "udevadm settle",
    +      )
    +      print(machine.succeed("find /dev/disk/ '!' -type d -printf '%p → %l\n' | sort"))
    +      machine.succeed(
               "mkswap ${grubDevice}-part1 -L swap",
               "swapon -L swap",
               "mkfs.ext3 -L nixos ${grubDevice}-part2",
               "mount LABEL=nixos /mnt",
               "mkdir -p /mnt/tmp",
           )
    +      machine.succeed("echo success")
    +      machine.succeed(
    +          'if [[ "$(find ${grubDevice} -printf \'%l\')" != "../../sdb" ]]; then exit 22; else true; fi'
    +      )
         '';
         grubVersion = 1;
         # /dev/sda is not stable, even when the SCSI disk number is.

And ran this way:

     $ until (clear; tmux clear ; time env -i nix-build nixos/release-combined.nix -A nixos.tests.installer.grub1.x86_64-linux); do echo derp; done

0 files changed, 0 insertions, 0 deletions