Document security guarantees & threat model in the README

We've received this feedback: "It would be very helpful if you could outline a scope or threat model in the Security section of the README since it is not obvious what the security guarantees are that Bazel offers." and questions around what constitutes a security issue in Bazel and what not occasionally come up.

As an example, Bazel, just like any other build system, of course runs a lot of code from various sources as part of its normal operation. The fact that someone can write a potentially malicious `genrule` that will be executed as part of a build is probably "working as intended" (just like a compiler or integration test could do harmful stuff). Bazel's sandbox is (currently) not meant to be a security boundary.

On the other hand, a hypothetical bug like "Bazel ignores the SHA256 of a downloaded http_archive" would be pretty bad.

We should document this for the benefit of our users and security researchers and to avoid potential misunderstanding about perceived security issues.