Error message about distributed deadlock when index name too long

We hit a similar case as is tested in the regression tests, with an error like:

```
BEGIN;
SELECT count(*) FROM test_index_creation1;
count
---------------------------------------------------------------------
0
(1 row)

CREATE INDEX ix_test_index_creation4
ON test_index_creation1 USING btree
(tenant_id, timeperiod);
ERROR: The index name (test_index_creation1_p2020_09_26_xxxxxx_tenant_id_timeperiod_idx) on a shard is too long and could lead to deadlocks when executed in a transaction block after a parallel query
HINT: Try re-running the transaction with "SET LOCAL citus.multi_shard_modify_mode TO 'sequential';"
```

I get that this behavior is apparently expected, but I couldn't find any information on the "why". How does an index name affect distributed deadlocks?