Add a minimal Pydantic model for Mimir coordinator config

[Moetez-Fradi]

Type-check existing fields so invalid config fails at build time instead of sending broken data over relations.

Issue

The coordinator builds Mimir config as an un-typed dict and ships it over relation data, so a malformed config only fail later, when workers try to start (long retry loop). See #38

This PR is the first part requested in the review: a minimal Pydantic model of the fields we already emit (tempo-style).

Solution

• Add nested Pydantic models in coordinator/src/mimir_config.py covering only the sections MimirConfig.config() already produces (alertmanager, ingester, limits, storage, TLS, etc.).
• Validate the assembled dict with MimirConfigModel.model_validate(...) before yaml.dump, so a bad shape fails on the coordinator at build time.
• Keep _build_* helpers returning dicts so existing unit tests stay intact.
• Use extra="ignore" so unknown keys (e.g. 2.17 ↔ dev) are dropped rather than raising. S3 inner keys stay Dict[str, Any] cause provider/track field names differ.

Context

Same pattern as Tempo: https://github.com/canonical/tempo-operators/blob/main/coordinator/src/tempo_config.py

Pydantic is already in the coordinator environment via charm libs / coordinated-workers; this change only uses it for the config we generate.

Testing Instructions

From coordinator/:

tox -e unit
tox -e lint
tox -e static

Or from the repo root: tox -e unit.

New cases in coordinator/tests/unit/test_mimir_config.py:

• round-trip with no S3 / no TLS
• unknown extra keys ignored
• S3 + TLS still validates (storage_prefix replaces filesystem)
• invalid types raise ValidationError

Upgrade Notes

charm config, relation, and storage have not been changed. Config YAML keys/values are aslo unchanged for current fields.

Unknown extra keys are ignored so a newer track adding a field this model does not know about should not fail validation. Older payloads still parse cause new model fields are optional or have defaults.