[leap5] in undo tests, create undo_index in CB memory segment and remove removed_nodes_tracker

[spoonincode]

Creating the undo_index on the stack (solely done in the undo tests) ultimately causes the index to contain a mixture of stack allocated and shared-memory allocated objects. Unfortunately I don't have an exact proof of how this happens but debugging through the structure's node_ptrs makes it quite clear it's occurring.

On Linux this typically is not a problem because the stack and shared-memory region tend to be very close to each other due to the implementation detail of where the mmap() is placed. However, on macOS, mmap()s get fulfilled from the "other direction": the stack will be at a high address where the mmap() will be placed at a very low address. This breaks the new compressed node_ptr structures and causes the test to crash.

It's possible to simulate this on Linux by forcing the chainbase shared memory to be located further away from the stack by gobbling up some VM space early, for example, just this line below and then run a simple test like the test_insert_fail will also crash on Linux.

diff --git a/src/pinnable_mapped_file.cpp b/src/pinnable_mapped_file.cpp
index a78cc6a..6786698 100644
--- a/src/pinnable_mapped_file.cpp
+++ b/src/pinnable_mapped_file.cpp
@@ -76,6 +76,7 @@ pinnable_mapped_file::pinnable_mapped_file(const std::filesystem::path& dir, boo
    _writable(writable),
    _sharable(mode == mapped)
 {
+   void* wut = mmap(nullptr, 16ull*1024*1024*1024*1024, PROT_NONE, MAP_PRIVATE | MAP_ANON, 0, 0);
    if(shared_file_size % _db_size_multiple_requirement) {
       std::string what_str("Database must be mulitple of " + std::to_string(_db_size_multiple_requirement) + " bytes");
       BOOST_THROW_EXCEPTION(std::system_error(make_error_code(db_error_code::bad_size), what_str));

Placing the undo_indexs in the shared memory region avoids the creation of mixed node_ptrs.

Unfortunately there still remains a similar issue in removed_nodes_tracker. I was unable to rectify this. But... removed_nodes_tracker is only used for a couple tests. I cannot find any harm in simply removing it?

fwiw a run on Leap with this branch is at,
https://github.com/AntelopeIO/leap/actions/runs/7493988034
as more clear evidence the removed code is unused.