What's the motivation for this PR?
I think we'll need to upgrade at some point, but there are some concern:

• We want to make things buildable with bazel 1.0+ so we've been using bazel 1.0 for our builds to make sure that works. The reason why we don't want to only guarantee build with the lastest bazel major is that we don't want to impose bazel version on 1. our dependencies (some of them might not have migrated to newer bazel version yet) 2.) our users (they should be able to use grpc in their project without having latest bazel).
• We currently don't have distribtests for testing buildability with different bazel versions, so we should not upgrade bazel until we have some (Ideally we'd test build with bazel 1.0 and 2.0).

@gnossen owns the KR to add bazel distribtest this quarter.

Some other related PRs:
Toolchain upgrade: #23087
Draft for distribtests: #21668