Testcase and fix for unicodeToAscii7() underallocation bug
This is an alternative solution to the one proposed in !327 (which uses a
len * 8 allocation size).
This MR's keeps using a
len * 2 allocation size, but now in a safe manner as we cap each converted Unicode char to a maximum of 2 ascii chars (which is good enough for our needs wrt text search functionality).
Includes a testcase which reproduces the bug.