also simplifies the code and now exactly matches postgresql search rules

I am not sure if there is a rationale behind ignoring extra keys like
this. I was under the impression that something might copy options to
platform specific options, thus leaving options in the original array,
knowing they would be ignored anyway. If it is the case, it would be
best to just move them, otherwise, this commit might help troubleshoot
some mapping issues.
I had to fix some tests that wrongly used that options array (maybe
these options were valid at some point?).

- SERIAL and BIGSERIAL are not true types and can't be used in ALTER TABLE expressions
- When changing between int and bigint auto increment fields the default should no be dropped
  to preserve the link between the column and the sequence.

Postgres has a `SERIAL` data type to allow shorthand for
`nextval('sequence_name')`, but DDL like this:

    CREATE TABLE example (id SERIAL)

Cannot be generated in the DBAL schema APIs.

And then we don't need to wait till v3 to not require comments on the
JsonType.

Features and changes:
* jsonb can be used with: options={"jsonb"=true}
* OVER is no longer reserved
* LATERAL is now reserved

Because PostgreSQL will return the expression with a lot of
parentheses we cannot TRIM() them easily. It is easier and more
correct to adapt to what PostgreSQL returns. That means the declaration
of partial indexes must be updated as follow:

Before:
``options={"where": "other_id IS NULL"}``

After:
``options={"where": "(other_id IS NULL)"}``

And fore more complex conditions, that would be:
``options={"where": "(((id IS NOT NULL) AND (other_id IS NULL)) AND (name IS NULL))"}``

Change name method from "getSqlDeclaration" to "getSQLDeclaration".

The AbstractSchemaManager expects getting the regex with its
delimiters since 8beb732f but the tests were not updated.