Expose detector bank totals to NICOS as derived devices

[SimonHeybrock]

NICOS asked for detector counts in the derived-device lists of all instruments, alongside the monitor counts they already read. Each detector bank now exposes its cumulative total as {source_name}_counts_total on {instrument}_livedata_nicos_data, under the same naming convention and the same start_time generation marker as the monitor devices (ADR 0006).

No backend or wire changes were needed: DeviceExtractor runs in every service's OrchestratingProcessor and the topic routing already exists, so the detector_data service starts publishing the moment the contract designates one of its outputs. The change is a registry declaration plus the regenerated exports.

Designation. Monitors have a single monitor_histogram spec per instrument, so their device is unambiguous. A detector bank is typically covered by several views — a projection, a strip view, a tube view — that all produce counts_total_cumulative, and device names must be unique across an instrument. Each instrument therefore designates exactly one view per bank, generally the primary geometric or whole-bank image. Designating a second one for the same source fails loud at contract construction. The shared COUNTS_TOTAL_DEVICE constant keeps the naming convention in one place for monitors and detectors alike, and its docstring carries the uniqueness rule.

Consequence worth knowing. The designated view becomes device-bearing, so the dashboard's confirmation gate now intercepts reset/stop/reconfigure on the primary detector view of every instrument while a job is running, naming the affected devices. Previously only monitor_histogram was gated.

Not fixed here: counts_total is the sum of the accumulated histogram, so events outside the view's TOA or wavelength edges are excluded. This is pre-existing and already applies to the monitor devices; in the default TOA configuration the edges span a full pulse period and nothing is dropped, but a narrowed range or wavelength mode truncates. Raised as #1242 for ESSlivedata and NICOS to settle.

Test plan

• A new test asserts every monitor and detector bank of every instrument has a counts device, so a new bank or a new instrument cannot silently miss one. Verified it fails when a designation is removed.
• A new service-level test asserts a running detector view publishes its bank total on the NICOS stream under the contracted device name, carrying start_time and end_time.
• Existing detector service tests now filter to the result stream, matching what the monitor service tests already do.
• The committed device_contract.yaml exports are regenerated; the existing drift test guards them.

[SimonHeybrock]

LGTM