configs: *: Restrict non-RVA23 CPU errata to rva20

[dramforever]

These extra CPU errata handling don't apply to RVA23 machines, and a few of them introduce noticible overhead by inserting nops in critical paths. Disable all of them.

[Icenowy]

Whether newer cores need CONFIG_ERRATA_THEAD_PMU seems to be not very clear?
For other workarounds it's okay, they're for non-A23 cores.

[RevySR]

Whether newer cores need CONFIG_ERRATA_THEAD_PMU seems to be not very clear? For other workarounds it's okay, they're for non-A23 cores.

The C908/C907/C920v2+ all have svpbmt.

[Icenowy]

Svpbmt isn't related to PMU -- it's related to MAE.

[RevySR]

Svpbmt isn't related to PMU -- it's related to MAE.

You are right, the SG2044 needs this PMU.
I remembered it wrong. The SG2044 does not support cycle/inst overflow interrupts.

[dramforever]

@Icenowy CONFIG_ERRATA_THEAD_PMU only applies to cores with marchid = 0 and mimpid = 0, which is older cores:

https://github.com/torvalds/linux/blob/8d3ae59288f1e7d58d76558a6ee96d533bc5019f/arch/riscv/errata/thead/errata.c#L130-L144

At least on SG2044 (which is not quite RVA23), it has (standard) Sscofpmf and non-zero marchid and mimpid. I think it is safe to assume all RVA23 "thead" cores would not need this. Even if SG2044 needs this, since this condition means it doesn't apply on SG2044 anyway, this wouldn't be a functional regression.

The same goes for CMO and MAE