Modules can report why they failed

A module had one channel back to its caller: the int returned from
dpm_module_execute, handed through by dpm_execute. Any detail behind
that number could only reach a log, so a caller wanting the reason had
to read output rather than ask for it.

dpm_set_last_error joins the exported API. A module records its reason
on the context handed to its entry point, which belongs to the caller,
and the caller reads it back with dpm_get_last_error.

The two accessors on the context now say what they do to it:
dpm_module_path becomes dpm_get_resolved_module_path, naming the value
it reports rather than the setting it came from, and dpm_last_error
becomes dpm_get_last_error, pairing with the setter.

Path normalization moves to sanitizers.cpp, which holds the conversions
that put a value written by a person into the single form the library
stores it in.

src/cli/dpm.cpp

@@ -112,7 +112,7 @@ namespace {
     int list_modules(dpm_ctx* ctx) {
         dpm_cursor* cur = dpm_list_modules(ctx);
         if (!cur) {
-            const char* err = dpm_last_error(ctx);
+            const char* err = dpm_get_last_error(ctx);
             std::fprintf(stderr, "dpm: %s\n", err ? err : "module listing failed");
             return 1;
         }
@@ -125,7 +125,8 @@ namespace {
         dpm_cursor_free(cur);
 
         if (infos.empty()) {
-            std::printf("No valid modules found in %s\n", dpm_module_path(ctx));
+            std::printf("No valid modules found in %s\n",
+                        dpm_get_resolved_module_path(ctx));
             return 0;
         }
 
@@ -250,7 +251,7 @@ int main(int argc, char** argv) {
 
     dpm_module* mod = dpm_require(ctx, module_name);
     if (!mod) {
-        const char* err = dpm_last_error(ctx);
+        const char* err = dpm_get_last_error(ctx);
         std::fprintf(stderr, "dpm: %s\n",
                      err ? err : "module could not be loaded");
         dpm_close(ctx);