+ * Parse a token from the input text. The modes are:
+ * MODE_QUOTED: Use *ep->start as the delimiter; the token ends
+ * before its next occurence. Do not interpret the token in any
+ * way and return EQN_TOK_QUOTED. All other modes behave like
+ * MODE_QUOTED when *ep->start is '"'.
+ * MODE_NOSUB: If *ep->start is a curly brace, the token ends after it;
+ * otherwise, it ends before the next whitespace or brace.
+ * Do not interpret the token and return EQN_TOK__MAX.
+ * MODE_SUB: Like MODE_NOSUB, but try to interpret the token as an
+ * alias created with define. If it is an alias, replace it with
+ * its string value and reparse.
+ * MODE_TOK: Like MODE_SUB, but also check the token against the list
+ * of tokens, and if there is a match, return that token. Otherwise,
+ * if the token matches a symbol, return EQN_TOK_SYM; if it matches
+ * a function name, EQN_TOK_FUNC, or else EQN_TOK__MAX. Except for
+ * a token match, *ep->start is set to an allocated string that the
+ * caller is expected to free.
+ * All modes skip whitespace following the end of the token.