low performance on insensitive case

Describe the bug
There's a performance Bug in the filters __iendswith, istartswith, iexact, icontains

To Reproduce
I have a table with 24M records,

class Company(Model):
    vat_number = fields.CharField(20, pk=True)
    name = fields.CharField(200, null=True)
    creation_date = fields.DateField(null=True)
    legal_category = fields.CharField(10, null=True)
    business_activity = fields.CharField(10, null=True)
    headquarters_nic = fields.CharField(5, null=True)
    created_at = fields.DateField(auto_now_add=True, null=True)

    class Meta:
        indexes = [
            GinIndex(fields=("name",))
        ]

when i use:
companies = await Company.filter(name__istartswith="something)

it takes about 25 seconds to retrieve data from the database

after checking the generated sql

print(await Company.filter(name__istartswith="something).as_query())

# SELECT "headquarters_nic","name","creation_date","created_at","vat_number","business_activity","legal_category" FROM "company" WHERE UPPER(CAST("name" AS VARCHAR)) LIKE UPPER('something%') ESCAPE '\' LIMIT 50

so the problem is the UPPER(CAST("name" AS VARCHAR)) if sql makes upper to all records to make a compare, it will take a lot of time.

Expected behavior
the solution, at least for postgres is to generate the follow sql:

SELECT "headquarters_nic","name","creation_date","created_at","vat_number","business_activity","legal_category" FROM "company" WHERE CAST("name" AS VARCHAR) ILIKE something%' ESCAPE '\' LIMIT 50

with this option the time to retrieve a response passed to about 300ms

Additional context

Start from the Company.filter(name__istartswith=...) entry point and trace how the ORM generates SQL for insensitive filters. Compare the PostgreSQL query shown in the issue with the expected ILIKE form, then identify the relevant tests or database-specific query path. Done means the PostgreSQL filters avoid the reported UPPER expression without breaking other database backends.