…2177)

Migrate the ad-hoc model-string sniffing for extended DSP and diversity to
the FlexLib-sourced ModelCapabilities table (Constitution Principle I —
FlexLib is the model authority), and fix the gaps that sniffing left behind.

Root of the reported AU-510 bug (extended DSP filters NRL/NRS/RNN/NRF not
appearing) is two-fold:

- Classification: the old hasExtendedDspFilters() used substring checks
  (contains("ML-")/"CL-"/"AU-") that silently missed the "S" server
  variants (MLS-9601 has no "ML-"; CLS-9301 has no "CL-") and was
  case-sensitive. Now derived from RadioPlatform (BigBend|DragonFire) via
  the table, which enumerates all 23 current FlexLib models.
- Timing: setHasExtendedDsp() was only pushed once at onSliceAdded, so if
  the `model` status arrived after the slice (GUIClientID session restore)
  the flag stayed stale. It is now also refreshed on RadioModel::infoChanged
  alongside SmartSDR+/diversity.

Diversity was the same anti-pattern, triplicated across three call sites and
wrong in both directions vs FlexLib IsDiversityAllowed:
- FLEX-6500 (single-SCU) was wrongly granted diversity.
- ML-/MLS-/CL-/CLS- (dual-SCU) were wrongly denied it.
All three sites now use RadioModel::isDiversityAllowed(), table-backed.

Also: RT-2122 (DragonFire, SliceList {A,B}) now reports 2 slices instead of
falling through to the default 4.

ModelCapabilities gains a RadioPlatform enum, a platform field, an
isDiversityAllowed flag, and a hasExtendedDsp() helper; the model table is
corrected (ML-380/CL-200/RT-100 placeholders -> real ML-9600/CL-9300/RT-2122)
and completed (MLS-9601, CLS-9301, AU-510). model_capabilities_test asserts
platform + 4m/2m/loop + diversity + extended-DSP for all 23 models, with
explicit regressions for the MLS/CLS/AU-510/6500 cases.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…atform field

Adding the RadioPlatform `platform` field as the first member of
ModelCapabilities broke xvtr_policy_test's positional brace inits
({false,true} / {true,false} assumed the old {has4Meters, has2Meters, ...}
order), which failed to compile in CI (the local incremental build hadn't
rebuilt this target). Switch to designated initializers ({.has2Meters=true}
/ {.has4Meters=true}) so the call sites are order-independent and resilient
to future capability fields.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

…orm table

Complete the single-source-of-truth migration started for extended-DSP and
diversity: slice/panadapter capacity and firmware-family routing now derive
from the FlexLib-sourced ModelCapabilities table (Principle I) instead of
parallel model.contains() ladders that had drifted.

- ModelCapabilities gains a maxSlices field (FlexLib SliceList size; also the
  max panadapter count, since pan capacity tracks SCU/slice capacity on every
  current model). Default 2 mirrors FlexLib's DEFAULT entry.
- maxSlicesForModel() and maxPanadapters() now return capabilitiesFor().maxSlices.
  Fixes the dual-SCU ML-9600/MLS-9601/CL-9300/CLS-9301: maxPanadapters() omitted
  them from its contains() list, capping them at 2 panadapters instead of 4.
- FirmwareStager::modelToFamily() routes by RadioPlatform (DragonFire -> 9600,
  everything else -> 6x00) instead of contains("9600"). The old substring test
  misrouted both ways: ML-9600/ML-9600W (BigBend) went to the government 9600
  firmware family, while RT-2122 (the actual DragonFire radio) fell to 6x00.

Not migrated (no FlexLib source): maxPanBandwidthMhz/minPanBandwidthMhz come
from packet captures (#1385) and FlexLib has no bandwidth field; hasGpsHardware
isn't in ModelInfo either. Left as-is rather than guessing ML/CL values.

model_capabilities_test asserts maxSlices for all 23 models plus explicit
ML/CL=4 and AU-510M=2 / AU-520M=4 regressions. Full app + test build verified.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>