Simplify Revlog->FSRSItem conversion (#13)

* Add comment about the required file to the test; move into test module

* Add English translations to extract_time_series_feature()

* review_kind = 4 is excluded in SQL query

* Shift review_kind=3 check into SQL

* Detect non-learning starts earlier

.retain() is fairly expensive, as it requires shuffling data about; better
to filter the material before the array is created

* Add some more checks

* Convert Vec<Revlog> directly to FSRSItem

* Review->FSRSReview

* Make label/delta_t the same type as FSRSReview; do conversion later

* Add a test to check tensor conversion

* Fix clippy lints

* Move current review into the list of reviews, instead of storing separately

* Remove some unused clones

Cargo.toml

@@ -14,3 +14,4 @@ log = "0.4"
 rusqlite = { version = "0.29.0" }
 chrono = "0.4.26"
 chrono-tz = "0.8.3"
+itertools = "0.11.0"