Selector rewrite: values other than long, interpret identifiers

according to context, to make `lstat * ?type#spy&spy>1' work.  Closes
bug#825363, #905809, #905814 and #922968.
(nsc_type, packed_nsc_type, nsc_cat, packed_nsc_cat, nsc_flags): New.
(valstr): New.  Old code encoded values in type long, which was
somewhat hard to read and could only support signed integer values.
(nscstr): Redesign.  Use valstr.  Typed operator.
(castr): Split ca_code into ca_type, ca_flags, ca_off.  Tables
changed.
(nstr_comp, nstr_exec): Redesign and rewrite.  Callers changed.  They
used the old design incorrectly, which let players smash the stack
by supplying more than NCOND conditions.
(encode, nstr_comp_val, decode, nstr_exec_val): Rename, redesign, and
rewrite.  Callers changed.
(nstr_coerce_val): New.
(var_ca, sect_ca, ship_ca, land_ca): Checking both var_ca[] and the
object's ca complicates proper recognition of unique abbreviations.
Copy contents of var_ca[] into the ca of objects, remove var_ca[].

(surv): Reject values with category other than NSC_OFF and types that
can't be coerced to NSC_LONG.  Old code happily passed values with
category NSC_VAL to code_char().  The previous version interpreted
them correctly, but earlier versions interpreted them as NSC_OFF, then
logged `bad type in decode: 0' and evaluated them into zero.

(code_char): Used to test category NSC_VAR to decide whether to
display tens or hundreds.  NSC_VAR no longer exists.  Test type
instead.  Makes more sense anyway.

src/lib/commands/sect.c

@@ -111,7 +111,7 @@ sct(void)
 	    continue;
 	ptr = &map[ns.dy][ns.dx];
 	*ptr = dchr[sect.sct_type].d_mnem;
-	if (nstr_exec(cond, ncond, (s_char *)&sect, EF_SECTOR)) {
+	if (nstr_exec(cond, ncond, &sect)) {
 	    ++nsect;
 	    *ptr |= 0x80;
 	}