Flaky store tests: CreateDevice test helper generates colliding MACs under the accepted-mac unique index

[gustavosbarreto]

Summary

TestPgStore/StatsStore (and any store test that creates more than one accepted
device in the same namespace) is flaky. It fails intermittently with:

--- FAIL: TestPgStore/StatsStore/succeeds_with_specific_tenantID
    stats_tests.go:49:
        Error Trace: api/store/storetest/helpers.go:226
                     api/store/storetest/stats_tests.go:49
        Error: Received unexpected error:
               document duplicate

Seen on CI: https://github.com/shellhub-io/shellhub/actions/runs/28873799116/job/85643304102
A plain re-run of the same commit goes green, which is the tell-tale of a flake.

Root cause

CreateDevice in api/store/storetest/helpers.go builds the default device MAC
from the wall clock:

Identity: &models.DeviceIdentity{MAC: fmt.Sprintf(
    "%02x:%02x:%02x:%02x:%02x:%02x",
    time.Now().UnixNano()%256, time.Now().UnixNano()%256, ...)},

On a runner whose clock resolution is coarser than the six time.Now() calls,
all six octets collapse to the same value, so the whole MAC becomes
VV:VV:VV:VV:VV:VV. Two devices created back to back then collide on their MAC
with probability ~1/256 per pair.

That collision was harmless until migration 007_devices_accepted_mac_unique
(#6600) added a unique index on (namespace_id, mac) WHERE status = 'accepted'.
Now a colliding MAC across two accepted devices in the same namespace turns into
a document duplicate on insert. StatsStore/succeeds_with_specific_tenantID
creates three accepted devices in tenant1, so it is the most frequent victim.

The default UID (fmt.Sprintf("%064x", time.Now().UnixNano())) has the same
weakness: two devices created within one clock tick collide on the primary key.

Suggested fix

Give the test helper collision-free defaults instead of deriving them from the
clock. A package-level atomic.Uint64 counter that feeds both the UID and the
MAC is enough:

var deviceSeq atomic.Uint64

n := deviceSeq.Add(1)
uid := fmt.Sprintf("%064x", n)
mac := fmt.Sprintf("00:00:%02x:%02x:%02x:%02x", byte(n>>24), byte(n>>16), byte(n>>8), byte(n))

Callers that need a specific MAC already pass one through a DeviceOption, so
only the default changes. The other clock-derived defaults in this file
(namespace name, user email, session UID) are candidates for the same treatment
if they start colliding under their own unique constraints.

Notes

This is a pre-existing issue in the test helper on master, exposed by the
#6600 unique index. It is not tied to any one feature branch; it can fail any PR
whose run happens to hit the collision.