docs: current-state refactor — merge scheduled-MAC contracts, drop milestone framing, separate lab-rig facts

[josephnef]

Documentation review + refactor across the tracked doc corpus (README, CLAUDE.md, docs/, tests/README). Four themes:

1. One scheduled-MAC doc instead of four milestone reports

docs/dl-departure.md, docs/beacon-grant.md, docs/ack-txreport.md and docs/ue-rx-attribution.md were titled "(M0 contract 1–4)" and written against the "5G-NR RAN epic" / "M1" / "M2". They are now one capability doc — docs/scheduled-mac.md — four measured per-generation contracts (submit→air guard time, dynamic beacon grants, hardware ACK/TxReport, per-UE RX attribution). All measurements, tables and verdicts kept verbatim; only the milestone voice is gone. Test-script headers and the IRtlDevice.h comment repointed; natural cross-links added from aggregation / time-distribution / multi-ap-cellular / ap-mode.

2. No WIP/milestone framing anywhere

• jammer-resilience loses "(M5)" and the "M6 … is follow-up work" bullet; fhss loses "…and what is next" / "is future work" — boundaries are stated as current-state facts.
• 8822e-quirks "## Untested" → "## Coverage" (states the characterized set: one BL-M8812EU2, rfe 21).
• adaptive-link-validation "Open questions (need hardware not currently available)" → "What this validation does not cover" (scope statements; the attenuator-rig design stays as methodology).

3. Lab-rig facts separated from general hardware behaviour

Rig hostnames, ssh targets, BDF/IOMMU specifics and "on this rig/bench" phrasing are generalized in CLAUDE.md, time-distribution, timing-accuracy, wfb-ng-tuning, bench-testing-near-field and tests/README — tracked docs now describe behaviour any developer can reproduce on their own hardware. (The rig specifics live in the local, untracked INVENTORY.md.) Product identifications — "Radxa X4 onboard Wi-Fi", adapter models in the benchmark table — stay: they are provenance, not lab state.

4. Gaps vs recent PRs filled

• CLAUDE.md now covers the whole beacon/TSF/AP surface (ReadTsf, StartBeacon/StopBeacon/UpdateBeaconPayload, PinBeaconTbtt, timesync/tdma demos) and aggregation/ACK (SetAmpduMode, SetAckResponder, tx.report, GetRxQuality vs cell::UeRxAttribution), plus src/cell/ in the architecture map — none of it was mentioned before.
• README: the 8812BU row gets its missing "5/10 MHz capable" marker (it rides the 8822BU path); the 8821AU row loses a stray "AC"; the binaries table gains tdma, timesync, doctor, pcieprobe; "Going deeper" is restructured into themed groups covering every doc — previously 13 of 32 docs were unreachable from the README (the four contract docs had zero incoming links anywhere).
• Stale src/ paths fixed in fused-fec.md (files moved to src/jaguar1/ long ago); driver-primer.md opens standalone instead of leading with its sibling.

Deliberately untouched: everything Kestrel (the out-of-scope lines are for the Kestrel PR to update when it lands); the RTL8821CE row is not marked 5/10 MHz — narrowband over PCIe is code-plausible but not bench-validated.

Verification

• Sweeps over the tracked corpus return clean: no M0..M9/milestone/epic/"future work"/"not yet"; no rig hostnames/"this rig"/"this bench" (the devourer-testrig mentions are the tests/setup_vm.sh default VM name, i.e. tool documentation).
• Every docs//tests//src//examples/ path referenced from the docs resolves in the tree; old filenames are gone repo-wide.
• cmake --build build -j + all 20 ctest selftests pass (only comments changed outside docs).

🤖 Generated with Claude Code