jaguar1(8814): match vendor TX-power parity (5G groups, PG default-map, BB-swing)

[josephnef]

Three RTL88xxAU TX-power divergences from the vendor reference (reference/rtl8812au), found while auditing #110. All three bring devourer's Jaguar1 TX-power path to exact parity with the upstream driver.

Fixes

1. 5G channel→group boundaries (classify_channel) now match upstream rtw_get_ch_group verbatim: gp3 = 60..98, gp4 = 100..106, gp0 from 16, gp13 to 253. Previously channels 84..98 resolved to the wrong group and read the wrong per-channel base index.
2. EFUSE default-PG-map fallback in LoadTxPowerInfo — port of the kernel's hal_load_pg_txpwr_info source chain (PG_DATA → IC_DEF → DEF). A base cell whose EFUSE byte is invalid (> txgi_max = 63, e.g. an unprogrammed 0xFF) is now filled from the per-chip default map (rtl881{2,4,21}a_pg_txpwr_def_info) then the generic map, instead of used raw (which would clamp to 63). Programmed EFUSEs are byte-identical to before → 8812/8821 unaffected.
3. BB-swing fixed-dB encoding (phy_get_tx_bb_swing_8812a): 0x05/0x0A → 0x55/0xAA so all four RF paths receive the requested −3/−6 dB swing (matches PHY_GetTxBBSwing_8814A); previously only path A did.

Scope / validation

• Builds all per-chip configs; ctest green.
• These are parity corrections; they do not change the ch36 index on a healthy chip. The 8814 ch36 under-drive reported in RTL8814AU: 5 GHz TX power drastically under-driven (~1.5 Mbps vs kernel ~58 Mbps) #110 was determined to be a USB Vbus-sag rail artifact, not a code defect: the working kernel driver programs the identical per-path index (path-A MCS7 @ ch36 = 14), and an on-air SDR measurement shows devourer radiates strongly at ch36. These fixes matter for channels 84..98 and for unprogrammed/partial EFUSEs.

Refs #110

🤖 Generated with Claude Code