revsetlang: use sysbytes() instead of blind encode()

Otherwise we will call str.encode() on Python 2, which is wrong.

sysbytes() does encode('utf-8') on Python 3. But the source is
guaranteed ASCII, so it shouldn't matter.

With this change, `hg` now runs with `HGUNICODEPEDANTRY=1` set.
However, several tests are failing.

Differential Revision: https://phab.mercurial-scm.org/D6050

mercurial/revsetlang.py

@@ -62,8 +62,8 b' symbols = {}'
 
 # default set of valid characters for the initial letter of symbols
 _syminitletters = set(pycompat.iterbytestr(
-    string.ascii_letters.encode('ascii') +
-    string.digits.encode('ascii') +
+    pycompat.sysbytes(string.ascii_letters) +
+    pycompat.sysbytes(string.digits) +
     '._@')) | set(map(pycompat.bytechr, pycompat.xrange(128, 256)))
 
 # default set of valid characters for non-initial letters of symbols