qa/suites/fs/upgrade/mds_upgrade_sequence: make suite fully role-based

Commit d2aa439db90 ("qa: list mons/mgrs for cephadm install") replaced the
host.a/host.b roles with explicit mon/mgr roles so that the shared
ignore-auth-warn-cephadm fragment (which targets mon.a) works, but left
'roleless: true' in the 0-from yamls and the host.a targets in the
cephadm.shell steps.

With roleless mode enabled, the cephadm task fabricates host-named mons
on top of the mon.a/b/c roles (5 mons across 2 hosts) and bootstraps
without --orphan-initial-daemons. cephadm bootstrap then applies its
default mon spec, which races with the explicit placement applied later
by the task: the mgr deploys the host-named mon on the second host with
the default ports (3300/6789) that the explicit placement assigned to
mon.b. mon.b then crash-loops with 'unable to bind monitor ...
(98) Address already in use', the monmap never reaches 5 mons, and
every job in the suite dies in cephadm bootstrap with 'reached maximum
tries (180) after waiting for 180 seconds'. The host.a shell targets
would also fail later since no host.a role exists any more.

Drop 'roleless: true' so the suite uses the normal role-based cephadm
path (bootstrap with --mon-id/--mgr-id and --orphan-initial-daemons,
explicit mon/mgr placement, no default specs to race with), and retarget
the cephadm.shell fragments from host.a to mon.a, matching the other
role-based upgrade suites.

Fixes: https://tracker.ceph.com/issues/79685
Signed-off-by: David Galloway <david.galloway@ibm.com>