Thanks for working on this, Bzlmod migrations always seem to be a bit harder than expected, and this is no exception.

This is a large change, but all looks good. //go/tools/gopackagesdriver:gopackagesdriver_test is failing on a few platforms, but as soon as that's green, I think this is ready to go.

Replying to a couple things from Slack:

Flipped the order of toolchain registration to prefer from_file over download tags. Otherwise our old compatibility SDK would be used for go mod download which is no good. This feels good in general, use the go.mod toolchain by default unless user explicitly asks for a different version

This may cause a user-visible change in functionality, but I expect most users want exactly one version of Go, so it probably won't be an issue in practice.

We got some minor version bumps in go.mod and grew a few extra minor deps. Given we already exported some big stuff like grpc I think it's not a huge deal, but it would be nice in the future if we used isolated = True to avoid polluting the overall closure with our deps from //tests

The changes here look fine, this shouldn't have much impact on downstream users.

we still test with bazel version 6.5, shall we bump minimum to 7.x? (Do we want to wait for 9 to actually ship first?)

Bumping to 7.x is probably a good idea at this point. The 9.x release should be very soon, and 6.x module functionality was pretty new at the time.