deps: bump libmdbx to 0.13.6

[shekhirin]

DO NOT MERGE: blocked until we test this PR and Pectra is on mainnet

---

This bump has breaking changes in cursor behaviour.

• Cursor no longer positions on the last entry in the table if seek_exact didn't find a matching key

  reth/crates/storage/db/src/implementation/mdbx/mod.rs

  Line 1380 in 85fcd03

  let _unknown = cursor.seek_exact(ShardedKey::new(real_key, u64::MAX)).unwrap();

  reth/crates/storage/db/src/implementation/mdbx/mod.rs

  Lines 971 to 975 in 1f958b4

  	assert_eq!(cursor.seek_exact(key2), Ok(None));
  	assert_eq!(
  	cursor.delete_current(),
  	Err(DatabaseError::Delete(reth_libmdbx::Error::NoData.into()))
  	);

• Deletion of the current key the cursor is positioned at moves the cursor forward, not back

  reth/crates/storage/libmdbx-rs/tests/cursor.rs

  Lines 330 to 339 in ef69ad0

  	assert_eq!(
  	cursor.get_current().unwrap(),
  	Some((Cow::Borrowed(b"key2" as &[u8]), Cow::Borrowed(b"val2" as &[u8])))
  	);
  	cursor.del(WriteFlags::empty()).unwrap();
  	assert_eq!(
  	cursor.get_current().unwrap(),
  	Some((Cow::Borrowed(b"key3" as &[u8]), Cow::Borrowed(b"val3" as &[u8])))
  	);

[erthink]

@shekhirin, Please use the stable branch (where is a regression hotfix and a few minors) instead of exactly the v0.13.5, or wait for the v0.13.6 release.

[shekhirin]

@erthink I see, we will wait for the 0.13.6, thank you! Do you have any estimates when do you plan to release it?

[erthink]

@erthink I see, we will wait for the 0.13.6, thank you! Do you have any estimates when do you plan to release it?

The 0.13.6 release is expected to be in 3 weeks, i.e. in April. But if you intend to use the 0.13.x branch, then I recommend not waiting for the next release, but achieving the tests now with the stable branch.

This will avoid discovering some kind of flaws immediately after the release (which just a scheduled in time manner).

[erthink]

@shekhirin, I just noticed that some cursors operation tests are broken.
I assume that the reason is not libmbx, but bindings or tests.

Since version 0.13, libmdbx has additional checks that prohibit operations with the cursor in a undefined/unpredictable state.
For instance, libmdbx prohibits relative-move or delete-current operations after an unsuccessful search, because a cursor position and/or state are not fully defined.

Thus a hidden heisenbugs are revealed.

--

Actually, of course, cursor(s) behavior are fully deterministic.
However, some search operations can be faulted/finished early/uncompelete, and therefore the result cursor position depends on the specific contents of the b-tree and its structure, which are unpredictable from the user's point of view when using the API.