Add indices on history table foreign keys

[pauljz]

This is only relevant for UUID columns, but results in some expensive joins. The way we build the GiST indices on history tables means we don't have any implicit index on the entity foreign key the way we do with the clock table. This adds an explicit index so you don't have to remember to cast your UUID primary keys to text when joining to column history.

---

Without casts. Note that the index scan only uses the vclock portion of the exclusion constraint and triggers an expensive join filter:

explain select * from my_table as t
left join trg_history.my_table_clock as c
  on t.id = c.entity_id
left join trg_history.my_table_history_my_col as hc
  on hc.entity_id = t.id and hc.vclock @> c.tick

Limit  (cost=0.70..2903.08 rows=100 width=1944)
  ->  Nested Loop Left Join  (cost=0.70..9313203.03 rows=320882 width=1944)
        Join Filter: (hs.entity_id = pa.id)
        ->  Index Scan using my_table_history_my_col_excl_vclock on my_table_history_my_col hc  (cost=0.28..16.97 rows=949 width=77)
              Index Cond: (vclock @> c.tick)

---

With casts! If we join on the UUIDs casted to text, we're back to being efficient:

explain select * from my_table as t
left join trg_history.my_table_clock as c
  on t.id = c.entity_id
left join trg_history.my_table_history_my_col as hc
  on hc.entity_id::text = t.id::text and hc.vclock @> c.tick

Limit  (cost=0.71..127.08 rows=100 width=1944)
  ->  Nested Loop Left Join  (cost=0.71..405509.70 rows=320882 width=1944)
        ->  Index Scan using my_table_history_my_col_excl_vclock on my_table_history_my_col hc  (cost=0.29..1.06 rows=1 width=77)
              Index Cond: (((entity_id)::text = (pa.id)::text) AND (vclock @> c.tick))

[multigl]

LGTM

[multigl]

tests run pretty fast here