Support for Match All index

[tomijaga]

Possible enhancement.
Not high priority right now.
i think the main benefit of this type of index is to save space on creating new indexes for nested fields that have a similar path.

[tomijaga]

Index on a field present at the same level in multiple variants

let #ok(_) = collection.create_index(["version.?.code"]);

let #ok(results) = collection.search(
 ZenDB.QueryBuilder().Where("version.?.code", #eq(#Nat(123)))
)

// results == [... all the variants with code value of 123]

let #ok(results2) = collection.search(
 ZenDB.QueryBuilder().Where("version.v4.code", #eq(#Nat(123)))
)

// results2 == [... all the #v4 variants with code value of 123]

[tomijaga]

Comments from parent issue:
The third index is a bit more complex than the first two as we would need to store the field names of the missing fields - those replaced by the match-all question mark symbol - along with the value in the index.

Also note that any index with a match-all field should override any of the exact path indexes that have the same nested field path.

For example, the collection could have two indexes "version.v3.code" and "version.v4.code".
Adding a new index with a match-all field, "version.?.code" for the version variant should replace both of those indexes with just the one.

Ideally, it would merge the entries from both indexes and repopulate the index for the missing fields (in this example, it would repopulate the index with any entry that has a #v1 or #v2 variant).

Since parts of the index has to be repopulated, it makes sense to create a new async method, that's separate from the default synchronous create_index() method, as we need to split this process into multiple async calls to avoid hitting the instruction limit.

• create_match_all_index()