why spawn_strategy=standalone in tensorflow's bazelrc?

It is hard to say, the commit which introduce that change do not tell why it was done.

The default option, which enable the best available sandboxing is probably the safest approach during development. However the Tensorflow case is different: most of the users do not change anything in a code base, they just want to build already tested codebase for their PC. In that case sandboxing can be disadvantageous, because it affects build performance, and there are some historical issues on some uncommon system configurations