t9301: Fix testcase covering up a bug in fast-import's notes fanout handling

There is a bug in fast-import where the fanout levels of an existing notes tree being loaded into the fast-import machinery is disregarded. Instead, any tree loaded is assumed to have a fanout level of 0. If the true fanout level is deeper, any attempt to remove a note from that tree will silently fail (as the note will not be found at fanout level 0). However, this bug was covered up by the way in which the t9301 testcase was written: When generating the fast-import commands to test mass removal of notes, we appended these commands to an already existing 'input' file which happened to already contain the fast-import commands used in the previous subtest to generate the very same notes tree. This would normally be harmless (but suboptimal) as the notes created were identical to the notes already present in the notes tree. But the act of repeating all the notes additions caused the internal fast-import data structures to recalculate the fanout, instead of hanging on to the initial (incorrect) fanout (that causes the bug described above). Thus, the subsequent removal of notes in the same 'input' file would succeed, thereby covering up the bug described above. This patch creates a new 'input' file instead of appending to the file from the previous subtest. Thus, we end up properly testing removal of notes that were added by a previous fast-import command. As a side effect, the notes removal can no longer refer to commits using the marks set by the previous fast-import run, instead the commits names must be referenced directly. The underlying fast-import bug is still present after this patch, but now we have at least uncovered it. Therefore, the affected subtests are labeled as expected failures until the underlying bug is fixed. Signed-off-by: Johan Herland <johan@herland.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>