Update CI workflow and drop slices for compatibility

[oliversun9]

This PR

1. Update Go directive in tools/go.mod and go.work to 1.21. This is OK because tools is not a public library.
2. Update .golangci.yml to exclude gci for results.go, because it has trouble recognizing slices as a standard library (even with Go 1.21)
3. Update make lint-go and use it in ci.yaml, also add target make lint-go-fix.

[oliversun9]

Without ignoring it, running golangci-lint with --fix would update this file's import list to

    "cmp"
	"github.com/bufbuild/protovalidate/tools/internal/gen/buf/validate"
	"github.com/bufbuild/protovalidate/tools/internal/gen/buf/validate/conformance/harness"
	"slices"

[nicksnyder]

All-else-equal, my preference would be to be consistent across all our repos. Right now our template repo and connect-go do not use the golangci-lint-action anymore (preferring instead to simply run make lint in CI). Aside from abstract consistency, this ensures that there is one authoritative version of the version of golangci-lint that we are using.

Would doing it this way here impact CI times at all? (if so, how much)? If no or insignificant impact, can we make this consistent with the other repos?

[oliversun9]

All-else-equal, my preference would be to be consistent across all our repos. Right now our template repo and connect-go do not use the golangci-lint-action anymore (preferring instead to simply run make lint in CI). Aside from abstract consistency, this ensures that there is one authoritative version of the version of golangci-lint that we are using.

Would doing it this way here impact CI times at all? (if so, how much)? If no or insignificant impact, can we make this consistent with the other repos?

Using running make lint sounds good.

Here's a past run with make lint, where it needs to download dependencies to build the linter, the lint steps took 44s. In another, this step took 19s.

A run on this PR with the lint action, where the lint step took 24s. In another, it took 22s.

I think the difference is small, 20s in the worst case, and will update the PR.