regard some errors as fatal even if --keep_going supplied

### Description of the bug:

We use --keep_going in some cases to collect a full set of build errors - it can be quicker to fix a set of build errors in one go rather than one-by-one. However, for some error states (such as no space left on device) it isn't helpful to keep going, and we would prefer that bazel immediately stops.

This is a request to consider some errors as fatal and stop immediately, even when --keep_going is supplied.

### Which category does this issue belong to?

_No response_

### What's the simplest, easiest way to reproduce this bug? Please provide a minimal example if possible.

Run a build with no disk space and --keep_going

### Which operating system are you running Bazel on?

windows

### What is the output of `bazel info release`?

7.3.1

### If `bazel info release` returns `development version` or `(@non-git)`, tell us how you built Bazel.

_No response_

### What's the output of `git remote get-url origin; git rev-parse HEAD` ?

_No response_

### If this is a regression, please try to identify the Bazel commit where the bug was introduced with bazelisk --bisect.

_No response_

### Have you found anything relevant by searching the web?

_No response_

### Any other information, logs, or outputs that you want to share?

_No response_

Start by reproducing a Bazel build with no disk space and --keep_going on Windows, using the reported Bazel 7.3.1 context. Trace how --keep_going handles that failure and identify which error states should stop immediately. Done means fatal errors halt the build while ordinary build errors still support collection with --keep_going.

Written by the indexing model from the issue text.