Fix issue #4538 "'load' strips r1 references in mergeinfo": don't filter out references to r1 in mergeinfo during 'svnadmin load' and 'svnrdump load'.
r1 is not an invalid mergeinfo source revision, merely an unlikely one, and we don't treat it as invalid anywhere else.
This reverts r941438 which has been in place since v1.7.0. Note also that this commit didn't remove r0 mergeinfo, only r1, despite the claim in its log message.
* subversion/libsvn_repos/load-fs-vtable.c, subversion/svnrdump/load_editor.c (renumber_mergeinfo_revs): Don't filter out references to r1. Canonicalize mergeinfo, where previously the combination of sorting followed by filtering achieved canonicalization.
* subversion/svndumpfilter/svndumpfilter.c (adjust_mergeinfo): Remove a comment that said r1 is invalid.
* subversion/tests/cmdline/svndumpfilter_tests.py (filter_mergeinfo_revs_outside_of_dump_stream): Correct a comment that said r1 is invalid, written in r941408.
Fix the 'svnadmin load' part of issue #4476 "Mergeinfo containing r0 makes svnsync and svnadmin dump fail".
Make 'svnadmin load' accept mergeinfo containing r0 (with a warning) if the option to validate properties is not enabled. In that case that mergeinfo property will not be adjusted for revision numbers or paths. Add a test.
* subversion/include/svn_repos.h (svn_repos_notify_warning_t): Add a new value for invalid mergeinfo.
* subversion/libsvn_repos/load-fs-vtable.c (adjust_mergeinfo_property): New, factored out of ... (set_node_property): ... here. If adjusting the mergeinfo fails, return an error if validating properties, or give a warning and use the invalid value otherwise.
* subversion/tests/libsvn_repos/dump-load-test.c (test_dump_bad_props): New, factored out of... (test_dump_r0_mergeinfo): ... here. (test_load_bad_props, load_r0_mergeinfo_notifier, test_load_r0_mergeinfo): New. (test_funcs): Add the new test.