[Bug]: treat failure to run the linter differently from linter reporting violations

### What happened?

We put the exit code of the linter into a file, and treat that as success.

However https://eslint.org/docs/latest/use/command-line-interface#exit-codes for example shows that a linter can distinguish between an error-level violation of a lint rule, and an error that prevents the linter from running. For example, a missing `dep` on an `js_library(srcs = eslint.rc.cjs)` target will make a successful lint action, with exit code 2 and no report. The tool that interprets the data doesn't have much to go on, as the error was printed to stderr the first time the tool ran, and isn't repeated in a cache hit nor captured in the report file.

### Version

HEAD

### How to reproduce

_No response_

### Any other information?

_No response_

Start by tracing how the linter exit code is written to the file and how the cached report is interpreted, using the js_library target with eslint.rc.cjs as the reproduction context. Compare ESLint's violation and execution-failure exit codes, including the stderr behavior on the first run. Done means lint violations remain reportable while failures to run are preserved and distinguished on cache hits.

Written by the indexing model from the issue text.