[feature] object_properties are a BTreeMap instead of BTreeSet

[woshilapin]

Related to hove-io/ntfs-specification#24.

[pbougue]

This looks OK.

Although we might have silent behavior changes, but I'm not sure how to detect them all: check .insert() returns None inside insert_object_property?

I'm not sure that it's an issue either: did not find doc specifying how the conflict should be resolved.

EDIT:
Maybe here https://github.com/CanalTP/transit_model/blob/9d0afbd9957dcef17c66bae2a63bf4bcc53e5ddc/src/ntfs/read.rs#L563
do a entry().or_insert() ?

[woshilapin]

Some details about most of the comments. If I'm summarizing, the main concern is what do we want between these 2 behaviors:

• when a duplicate key is to be inserted, we discard it, first one should win
• when a duplicate key is to be inserted, we just insert it, last one should win

After some more discussions about that topic, we decided that it does not matter that much which one is chosen, as long as it is consistent. Therefore, since last one should win is the default strategy for BTreeMap, we'll keep it as it is, this will avoid some surprising syntax like

properties_map.entry(some_key).or_insert(some_value);

instead of the usually more expected

properties_map.insert(some_key, some_value);