…UILD and .bzl input parsing.

Internally BUILD and .bzl files are read as uninterpreted raw byte strings and
not as UTF-8, so we should write content back out the same way.  For example,
if a BUILD file contains:

```
...
notice = "Copyright © 2019 Acme LLC",
```

Bazel will store the copyright symbol as two distinct octets ([0xc2, 0xa9]).
When writing that with an action, the same two octets should be emitted.
The behavior fixed by this change is to not interpret each character as
a standalone Unicode code point to be encoded as UTF-8.

While this change is appropriate for strings read from BUILD files, there is still a discrepancy in handling file paths with non Latin1 characters on Windows file systems.  That is described in #11602.

While this change does produce an observable behavior change, I am considering it a bug fix rather than an incompatible change because the previous behavior was both wrong and also (as described in #11602) inconsistent across different OSes.

Closes #10174

RELNOTES:
Fix behavior of ctx.actions.write so content is written without an incorrect encoding to UTF-8.
See #10174 for details.
PiperOrigin-RevId: 318056290