ANALYZE FORMAT=JSON on the original dataset and query from MDEV-28073 (the branch is based on old 10.10 and doesn't contain the full fix MDEV-28852 yet, so the query spends a lot of time in optimization).

| {

  "query_optimization": {

    "r_total_time_ms": 1199082.822

  },

  "query_block": {

    "select_id": 1,

    "r_loops": 1,

    "r_total_time_ms": 38.21517253,

...

1 row in set (19 min 58.541 sec)

So, even the query optimization time alone (1199082.822), not counting the query block time, already exceeds the total execution time reported by the client (19 min 58.541 sec = 1198541).
The difference here is negligible, but the mere fact that it can occur raises a question about possible bigger discrepancies.

Faster achievable with the same data/query under optimizer_search_depth=15:

| {

  "query_optimization": {

    "r_total_time_ms": 23634.83577

  },

  "query_block": {

    "select_id": 1,

    "r_loops": 1,

    "r_total_time_ms": 0.099659582,

...

1 row in set (23.626 sec)