Mismatch in primary key types should not cause a failure on SQL writing

[OriHoch]

reproduction

• table schema:

{"fields": [
    {"name": "id", "type": "integer"},
    {"name": "date", "type": "string"}
],
"primaryKey": ["id"]

• write some rows to a PostgreSQL DB - but in the rows, have the id as type string (but numeric)
• data in DB is inserted correctly, and DB converts it to integer
• now, write the same rows again

expected

• provide an error message, warning about type mismatch
• or match rows in DB (as opposed to bloom)

actual

• writer.__prepare_bloom - collects all data as integers
• writer.__check_existing - compares with a string id - which doesn't match
• this causes insert
• when inserting, the DB matches the primary key (even though types don't match)
• error about duplicate key

[OriHoch]

see: frictionlessdata/datapackage-pipelines#77

[roll]

Thank for reporting @OriHoch (cc @akariv)