Feat/mssql python Add optional mssql-python backend, make pyodbc optional, and update docs / CI

[axellpadilla]

This PR introduces the new optional mssql-python connection backend while keeping the legacy pyodbc ODBC path available as an explicit extra.

Key changes:

• Add lazy-loaded backend support in sqlserver_connections.py
• Make both pyodbc and mssql-python optional extras in pyproject.toml
• Update sqlserver_credentials.py for backend selection
• Add/adjust unit tests for lazy import and backend behavior
• Update integration test matrix in integration-tests-sqlserver.yml to preserve coverage without increasing total run count
• Add devcontainer support for both backend extras in setup_env.sh
• Clarify installation and usage in README.md
• Add developer setup notes and dependency guidance to CONTRIBUTING.md
• Small support changes in test.env.sample, conftest.py, and uv.lock

Why this change:

• enables the modern mssql-python backend behind currently gated use_mssql_python: true
• Avoids requiring system dependencies unless a backend is explicitly selected
• keeps existing pyodbc behavior intact for legacy profiles
• improves documentation and local development setup

Testing:

• updated unit tests for both backend import behavior
• CI workflow preserves the same total run count by partitioning backends across Python versions
• devcontainer setup installs both pyodbc and mssql-python extras for local validation

[Benjamin-Knight]

Ok, its a big change so will take a while to review, I can see a few issues I think so I'll find some time at start providing feedback.

My main concern other than the actual code change is the install method for the adapter now changes as we have to specify the driver we want to use, that is a big change. This doesn't just affect us, DBT themselves will have the wrong install instructions.

https://docs.getdbt.com/docs/local/connect-data-platform/mssql-setup?version=1.12#prerequisites

[Benjamin-Knight]

Probably need some new tests as well but will look at those after these comments.

[axellpadilla]

@Benjamin-Knight I'm moving this to v1.11, I'm more interested in the other patches for v1.10 and its true we will need more time holding the rc1 pending until dbt docs are updated

[axellpadilla]

@Benjamin-Knight Kept pyodbc by default and already running pyodbc part on production so changed milestone to v1.10.1, this way we don't need to update upstream docs still, and can announce deprecation if we really start to mostly use the new driver.
All lazy imports behavior intact, this mostly makes a mandatory install of pyodbc by default even if we don't use it.
This should be squash merged after v1.10.0

[Benjamin-Knight]

@Benjamin-Knight Kept pyodbc by default and already running pyodbc part on production so changed milestone to v1.10.1, this way we don't need to update upstream docs still, and can announce deprecation if we really start to mostly use the new driver. All lazy imports behavior intact, this mostly makes a mandatory install of pyodbc by default even if we don't use it. This should be squash merged after v1.10.0

Ok, Its a big change so we will want to have a much longer period for a release candidate then we did with the 1.9.1 release to give people time to feedback but I think its good to go now that the default behaviour is still pyodbc.

Have you flipped the adapter into mssql driver mode for a large corpus of models?

[axellpadilla]

@Benjamin-Knight Kept pyodbc by default and already running pyodbc part on production so changed milestone to v1.10.1, this way we don't need to update upstream docs still, and can announce deprecation if we really start to mostly use the new driver. All lazy imports behavior intact, this mostly makes a mandatory install of pyodbc by default even if we don't use it. This should be squash merged after v1.10.0

Ok, Its a big change so we will want to have a much longer period for a release candidate then we did with the 1.9.1 release to give people time to feedback but I think its good to go now that the default behaviour is still pyodbc.

Have you flipped the adapter into mssql driver mode for a large corpus of models?

Going to on a couple days of pyodbc, current side effect is a speed gain on sqlfluff formatting because the lazy stuff