Respect SAW_IMPORT_PATH

[RyanGlScott]

Due to an oversight, the SAW_IMPORT_PATH environment variable (or, equivalently, the -i/--import-path command-line options) were being blatantly ignored by SAW. This fixes that oversight by restricting the API in SAWScript.Import to a single exported function (findAndLoadFile), which always consults SAW_IMPORT_PATH before attempting to load a file.

Fixes #2756.

[RyanGlScott]

16 tests now fail because of changes in error messages that look like this:

-    within "x" (unify009.saw:20:6-20:7)
+    within "x" (./unify009.saw:20:6-20:7)

This is because the default SAW_IMPORT_PATH is . (i.e., the current path), and the import paths get prepended to the overall file path when calling findAndLoadFile. I wonder if we should add a special case for . such that ./foo.saw is instead represented as foo.saw?

[brianhuffman]

I wonder if we should add a special case for . such that ./foo.saw is instead represented as foo.saw?

Yes, I think we should drop the leading ./. It is longer, noisier, and provides no extra information; we're better off without it.

[brianhuffman]

I think we should go ahead and fix the leading "./" thing; other than that everything looks good to me.

[RyanGlScott]

I opted to fix this by using System.FilePath.normalise, which collapses ./foo.saw to foo.saw, among various other touchups. I think this should suffice to restore the original error messages' formatting of file paths, and it might also make the formatting a bit prettier in other obscure cases on Windows. Let's see what the CI says.

[RyanGlScott]

Using normalise causes the CI to pass, but with one slight technicality: the Windows CI job that runs the integration tests now has more failing tests than it did before. This is because on Windows, normalise will normalize / path separators to \\, as seen in this failing test case:

  test2301:                            eval saw  -j 'D:\a\saw-script\saw-script\intTests\jars\bcprov-jdk16-145.jar;D:\a\saw-script\saw-script\intTests\jars\galois.jar;D:\a\saw-script\saw-script\intTests\jars\org.sat4j.core.jar' test0.saw

Unexpected test diffs.
If the new outputs are correct, update the reference outputs, but
please don't do so without thinking.

--- test0.log.good	2025-10-30 14:45:25.597637700 +0000
+++ test0.log	2025-10-30 14:49:15.594833800 +0000
@@ -1,2 +1,2 @@
- Loading file "test0.saw"
- Loading file "subdir/testsym.saw"
+ Loading file "test0.saw"
+ Loading file "subdir\\testsym.saw"

FAIL (0.71s)
    intTests\IntegrationTest.hs:185:
    expected: ExitSuccess
     but got: ExitFailure 1
    Use -p '/test2301/' to rerun this test only.

The Windows integration tests CI job is already allowed to fail, so this doesn't affect whether the overall CI passes or not. Still, it does mean that if we ever did want to make the integration tests all pass on Windows, this PR would give us more work to do. Would folks be OK with accepting this? I could foresee ways to make the test suite more robust to Windows-style path separators (e.g., by making the test-diff infrastructure less sensitive to path-separator differences), but it would take some work to implement this.

[sauclovian-g]

Go ahead and merge and I'll take care of it. It should let me remove some annoying hacks I had to add in the last run. (The prior state of things was that some paths came out with / and some with \ and some with \\, arbitrarily.)