Correct imports and improve generation logic in SQLModelGenerator

[touale]

In fact, I found that there are some problems in the code generated by sqlacodegen for sqlmodel, which is manifested in the introduction of redundant class CHAR and the mapped_column does not been import. The following is a detailed description of the problem:

(1) When encountering CHAR(36) and CHAR(36, 'utf8mb4_general_ci'), sqlacodegen will repeatedly import from different sources

from sqlalchemy import CHAR
from sqlalchemy.dialects.mysql import CHAR

The adjustment I made was to focus on sorted and high-priority imports, and to make a warn about duplicate imports.

for example:

WARN: Duplicate imports `{'CHAR'}`  are detected from the package `sqlalchemy.dialects.mysql` and will be filtered, 
which may cause abnormal behavior.

In fact, "from sqlalchemy import CHAR" supports general projects of multiple databases, while "from sqlalchemy.dialects.mysql import CHAR" is specifically for MySQL projects. If you project requires cross-database compatibility or you are not sure about the target database type, it is safer to use "sqlalchemy.CHAR".

(2) sqlmodel does not support mapped_column in the latest release(https://github.com/fastapi/sqlmodel/releases/tag/0.0.22), although someone has proposed a PR (fastapi/sqlmodel#1143), so Column should be used for the generation of sqlmodels for the time being

[touale]

#352 can be solved