opt: overly-pessimistic statistics estimate for index join on top of inverted index scan

Our statistics estimates for index joins on top of inverted index scans seem wildly pessimistic. An index join should always have a row count estimate = the child row count estimate, but when the child is an inverted index scan we see estimates > 1000x the child estimate. Here's a demonstration:

```sql
CREATE TABLE ijk (i INT PRIMARY KEY, j JSONB, k STRING, INVERTED INDEX (j));
INSERT INTO ijk SELECT i, json_build_object('foo', i), i::string FROM generate_series(0, 9999) AS s(i);
ANALYZE ijk;
EXPLAIN SELECT i FROM ijk WHERE j->'foo' = '3';
EXPLAIN SELECT k FROM ijk WHERE j->'foo' = '3';
```

The first explain, using only an inverted index scan, shows a row count estimate of 0 (which doesn't seem right, either, should be 1):

```
demo@127.0.0.1:26257/demoapp/defaultdb> EXPLAIN SELECT i FROM ijk WHERE j->'foo' = '3';
info
---------------------------------------------------------------------------------
distribution: local
vectorized: true

• scan
estimated row count: 0 (<0.01% of the table; stats collected 2 seconds ago)
table: ijk@ijk_j_idx
spans: 1 span
(7 rows)

Time: 1ms total (execution 1ms / network 0ms)
```

The second explain, with an index join on top of an inverted index scan, shows a row count estimate of over 1000 which is wildly pessimistic (the query only returns 1 row):

```
demo@127.0.0.1:26257/demoapp/defaultdb> EXPLAIN SELECT k FROM ijk WHERE j->'foo' = '3';
info
-------------------------------------------------------------------------------------
distribution: local
vectorized: true

• index join
│ estimated row count: 1,111
│ table: ijk@ijk_pkey
│
└── • scan
estimated row count: 0 (<0.01% of the table; stats collected 5 seconds ago)
table: ijk@ijk_j_idx
spans: 1 span
(11 rows)

Time: 1ms total (execution 1ms / network 0ms)
```

Jira issue: CRDB-28352