We recently fixed a long-standing bug with Subversion path

normalization, allowing some characters that weren't previously

supported to be properly escaped. However, this turned up some

regressions with Subversion 1.9 and older.
Subversion is very particular about which characters (with ASCII values

0 through 128) that must be escaped and which must not be. This doesn't

directly map to what urllib.quote() would normally escape.

Fortunately, it allows us to supplement the list of "safe" characters.
This change adds that list of safe characters based on the lookup table

in Subversion's code, and introduces new unit tests for comparing a URL

for the full 0-128 characters to the output that would be escaped using

exactly Subversion's table. It also runs a file existence check using

the same string, helping us ensure that we won't crash Subversion in CI

with an assertion check.

Tested this in a CI environment against Subversion 1.9, where the

crash was noticed. Unit tests passed.