pcie-brcmstb backports to 6.14/6.12

[P33M]

In linux-next, an upstream implementation of pcie-brcmstb has been posted, is nearly applied, and will be the baseline for support for BCM2711 and BCM2712 SoCs going forwards. There's a lot of implementation detail (and interop bugs) that this new-old driver needs to be taught about.

1. L1SS

• Does a L1SS-capable endpoint on a CM4/CM5/Pi5 work?
• Does hardware with CLKREQ# unconnected on CM4/CM5 work?
• Are the L1.2 PM timers correct on 2712?
• Does the driver attempt to enable SSC on 2712?

2. Outbound bridge

• Add in BCM2712 TLP completion error response suppression
• UBUS timeout increase on 2711/2712
• Check that timeouts on 2712 are set in PCIE_MISC registers not RGR1

3. BCM2712 QoS

• Add the "more magic" DT bindings
• Tailored VDM to QoS map for RP1
• Default QoS map for PCIEx1
• D0/C1 fixes present and correct (EN_QOS_UPDATE_TIMING_FIX et al)
• Your TLP delivered within 30 minutes or the next one is free? (Hailo/NVMe completion timeouts)

4. Read completion boundary hack

• RP1 doesn't care about the size of read completions, so put the DT wart back in
  Invalid, not needed - MPS completions are allowed, and PCIe rules for splitting completions bigger than MPS are satisfied by the RC.

5. Tperst_clk delay

• add DT wart back in

6. Upstreamed downstream fixes

• Is MDIO_WR_DONE still broken (no)
• CRSVis bit reset if set with perst_n active
• L1/L0s disable broken with aspm-no-l0s DT wart