monorail: Reconsider bf_sde workaround(s)

We currently have a workaround in monorail that determines if the network link between the tofino (main network switch) and vsc7448 (management network switch) does not properly autonegotiate:

https://github.com/oxidecomputer/hubris/blob/1cbf835c1db3c8d9a7a53274e5af38e002c1c244/task/monorail-server/src/bsp/sidecar_bcd.rs#L559-L598

In a few observed cases, CC cs#1485, this failure to autonegotiate is potentially due to the PCIe link between the scrimlet and tofino not operating nominally (and therefore not being a good autonegotiation partner for the vsc7448), rather than some unexpected behavior from the vsc7448 itself.

However, with this workaround, we completely re-sequence the vsc7448 every 20 seconds, meaning that the management network will be brought down and up again repeatedly until the link between the vsc7448 and tofino is resolved.

Although the link between the scrimlet and its management network won't be functional in this case (as it connects through the tofino, so no tofino <-> vsc7448 link means no access to the management network), we are also inhibiting tech-port access to the management network on that switch.

Unfortunately, as the scrimlet does not have any means of directly talking to its sidecar's SP, it's currently hard for us to know if the failure to negotiate is ephemeral (and we should keep retrying), or if it is more permanent (requiring a tofino re-sequence). We may want to make this negotiation process bounded, or with an increasing back-off.

Start with task/monorail-server/src/bsp/sidecar_bcd.rs around lines 559-598 and trace how the bf_sde workaround detects autonegotiation failure and re-sequences the vsc7448. Determine whether retrying should be bounded or use increasing back-off, then verify that the management network is not repeatedly disrupted while the tofino-to-vsc7448 link remains unresolved.