K2: Optimize AbstractFirDeserializedSymbolProvider

Avoid filling caches with keys that are definitely empty
(if it's cheap to compute that), to decrease the size of backing maps.

The strategy is pre-computing the sets of names that might be met.
NB: the size of the sets is way fewer than a size of all queried names.

analysis/low-level-api-fir/tests/org/jetbrains/kotlin/analysis/low/level/api/fir/services/PackagePartProviderTestImpl.kt

@@ -26,6 +26,9 @@ internal class PackagePartProviderTestImpl(
                 return providers.flatMapTo(mutableSetOf()) { it.findPackageParts(packageFqName) }.toList()
             }
 
+            override fun computePackageSetWithNonClassDeclarations(): Set<String> =
+                providers.flatMapTo(mutableSetOf()) { it.computePackageSetWithNonClassDeclarations() }
+
             override fun getAnnotationsOnBinaryModule(moduleName: String): List<ClassId> {
                 return providers.flatMapTo(mutableSetOf()) { it.getAnnotationsOnBinaryModule(moduleName) }.toList()
             }