FIX: Ensure proper path component types during refinement processing

A self-declared refinement processing "hack" fails, in certain
situations, to re-verify that path components accessed in refinement
processing are words before using word-specific accessor functions on
them.

Without this fix (example as reported by Sunanda in CC#1977):

    >> a: func [/b] [1]

    >> a/b/%
    ;; R3 crashes at this point.

With this fix:

    >> a/b/%
    ** Script error: incompatible refinement: %""

This patch is a slight variation of an earlier fix written by Shixin
Zeng (zsx/r3@72b5d98).

Fixes: Oldes/Rebol-issues#1977

(cherry picked from commit c3d6e02)

Author: earl

src/core/c-do.c

@@ -746,8 +746,10 @@ x*/	static REBINT Do_Args_Light(REBVAL *func, REBVAL *path, REBSER *block, REBCN
 	}
 
 	// Hack to process remaining path:
-	if (path && NOT_END(path)) goto more_path;
-	//	Trap2(RE_NO_REFINE, Func_Word(dsf), path);
+	if (path && NOT_END(path)) {
+		if (!IS_WORD(path)) Trap1(RE_BAD_REFINE, path);
+		goto more_path;
+	}
 
 	return index;
 }

src/tests/units/crash-test.r3

@@ -23,6 +23,11 @@ Rebol [
 	recycle ;@@ <-- it was crashing here
 	--assert 1 = a
 
+--test-- "issue-1977"
+	;@@ https://github.com/Oldes/Rebol-issues/issues/1977
+	a: func [/b] [1]
+	--assert error? try [a/b/%] ;- no crash, just error!
+
 ===end-group===
 
 ~~~end-file~~~