redfish: fix GPU duplication on Supermicro HGX, exclude NVSwitch, restore path dedup

Three bugs, all related to GPU dedup in the Redfish replay pipeline:

1. collectGPUsFromProcessors (redfish_replay.go): GPU-type Processor entries
   (Systems/HGX_Baseboard_0/Processors/GPU_SXM_N) were not deduplicated against
   existing PCIeDevice GPUs on Supermicro HGX. The chassis-ID lookup keyed on
   processor Id ("GPU_SXM_1") but the chassis is named "HGX_GPU_SXM_1" — lookup
   returned nothing, serial stayed empty, UUID was unseen → 8 duplicate GPU rows.
   Fix: read SerialNumber directly from the Processor doc first; chassis lookup
   is now a fallback override (as it was designed for MSI).

2. looksLikeGPU (redfish.go): NVSwitch PCIe devices (Model="NVSwitch",
   Manufacturer="NVIDIA") were classified as GPUs because "nvidia" matched the
   GPU hint list. Fix: early return false when Model contains "nvswitch".

3. gpuDocDedupKey (redfish.go): commit 9df29b1 changed the dedup key to prefer
   slot|model before path, which collapsed two distinct GPUs with identical model
   names in GraphicsControllers into one entry. Fix: only serial and BDF are used
   as cross-path stable dedup keys; fall back to Redfish path when neither is
   present. This also restores TestReplayCollectGPUs_DedupUsesRedfishPathBeforeHeuristics
   which had been broken on main since 9df29b1.

Added tests:
- TestCollectGPUsFromProcessors_SupermicroHGX: Processor GPU dedup when
  chassis-ID naming convention does not match processor Id
- TestReplayCollectGPUs_DedupCrossChassisSerial: same GPU via two Chassis
  PCIeDevice trees with matching serials → collapsed to one
- TestLooksLikeGPU_NVSwitchExcluded: NVSwitch is not a GPU

Added rule to bible-local/09-testing.md: dedup/filter/classify functions must
cover true-positive, true-negative, and the vendor counter-case axes.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

bible-local/09-testing.md

@@ -41,3 +41,41 @@ go test ./internal/exporter/... -cover
 - When adding a new vendor parser, include at minimum:
   - `Detect()` test with a positive and a negative sample file list.
   - `Parse()` test with a minimal but representative archive.
+
+## Dedup and filtering functions — mandatory coverage
+
+Any function that deduplicates, filters, or classifies hardware inventory items
+**must** have tests covering all three axes before the code is considered done:
+
+| Axis | What to test | Why |
+|------|-------------|-----|
+| **True positive** | Items that ARE duplicates are collapsed to one | Proves the function works |
+| **True negative** | Items that are NOT duplicates are kept separate | Proves the function doesn't over-collapse |
+| **Counter-case** | The scenario that motivated the original code still works after changes | Prevents regression from future fixes |
+
+### Worked example — GPU dedup regression (2026-03-11)
+
+`collectGPUsFromProcessors` was added for MSI (chassis Id matches processor Id).
+No tests → when Supermicro HGX arrived (chassis Id = "HGX_GPU_SXM_1", processor Id = "GPU_SXM_1"),
+the chassis lookup silently returned nothing, serial stayed empty, UUID was new → 8 duplicate GPUs.
+
+Simultaneously, fixing `gpuDocDedupKey` to use `slot|model` before path collapsed two distinct
+GraphicsControllers GPUs with the same model into one — breaking an existing test that had no
+counter-case for the path-fallback scenario.
+
+**Required test matrix for any dedup function:**
+
+```
+TestXxx_CollapsesDuplicates   — same item via two sources → 1 result
+TestXxx_KeepsDistinct          — two different items with same model → 2 results
+TestXxx_<VendorThatMotivated>  — the specific vendor/setup that triggered the code
+```
+
+### Practical rule
+
+When you write a new filter/dedup/classify function, ask:
+1. Does my test cover the vendor that motivated this code?
+2. Does my test cover a *different* vendor or naming convention where the function must NOT fire?
+3. If I change the dedup key logic, do existing tests still exercise the old correct behavior?
+
+If any answer is "no" — add the missing test before committing.