Another ckparity flaw

Found another ckparity flaw! Only detected now due to the reworked tag
encoding, but it's just luck this wasn't detected earlier.

Consider the following bit flip:

  03 04 80 04 80 04 6b ...  data w512 512
  43 04 80 04 80 04 6b ...  altble 0x304 w512 -512
  ^
  flip

Not only are leb128s problem for ckparity, but even the difference in
alt vs normal tag encoding presents a vulnerability.

So limiting the ckparity tests further, to just 47 of the first 48 bits.

tests/test_ck.toml

@@ -2110,6 +2110,11 @@ code = '''
             // to our revision count + first tag
             i < ((BADBIT == -1) ? 8*6 : 1);
             i++) {
+        // we can't even detect bit flips that change the
+        // alt vs tag encoding
+        if (BADBIT == -1 && i == 8*4+6) {
+            continue;
+        }
         lfs_size_t badbit = (BADBIT == -1) ? i : BADBIT;
 
         // reset the bd prng every run for reproducibility
@@ -2262,6 +2267,11 @@ code = '''
             // to our revision count + first tag
             i < ((BADBIT == -1) ? 8*6 : 1);
             i++) {
+        // we can't even detect bit flips that change the
+        // alt vs tag encoding
+        if (BADBIT == -1 && i == 8*4+6) {
+            continue;
+        }
         lfs_size_t badbit = (BADBIT == -1) ? i : BADBIT;
 
         // reset the bd prng every run for reproducibility