Only fail with IndexFailure on hasUncompilableErrorOccurred.

Summary:
There are a bunch of `IndexFailure` facts in the `fbsource` DB where 0 errors are captured. This is because we use the `hasErrorOccurred` as the condition for bailing, but we actually only care about error and fatals. `hasErrorOccurred` is true also for warnings propagated to error, and things like that. This diff uses `hasUncompilableErrorOccurred` instead.

```
fbsource> :count src.IndexFailure _

17159 results, 17159 facts, 29.64ms, 14761584 bytes, 584 compiled bytes

fbsource> :count src.IndexFailure { details = "0 error(s) occurred while indexing" }

13115 results, 13115 facts, 21.12ms, 6530968 bytes, 780 compiled bytes
```

Querying over the [Indexer Coverage](https://fburl.com/scuba/indexer_coverage/m5h3ylgh) table shows that fixing this issue will index ~3,500 more files directly in `fbsource`, and potentially more indirectly affected files. But the actual user impact will likely be lower, most of the 3,500 files are already indexed in other DBs. There are still [173 files](P829689600) that are not indexed in other DBs.

Reviewed By: malanka

Differential Revision: D49318003

fbshipit-source-id: f23e7ec351e553f8771a64c24e47e41505bae1df

glean/lang/clang/ast.cpp

@@ -2669,7 +2669,8 @@ struct ASTConsumer : public clang::ASTConsumer {
     // Clang is sometimes generating a bogus AST when there are
     // compilation errors (even for parts of the AST that should be
     // unrelated to the error) so this is a workaround.
-    if (ctx.getDiagnostics().hasErrorOccurred() && !FLAGS_index_on_error) {
+    if (ctx.getDiagnostics().hasUncompilableErrorOccurred() &&
+        !FLAGS_index_on_error) {
       db->IndexFailure(ctx);
       return;
     }