Expand the resolver seam to STAC collections/items (and MosaicJSON)

Expand the factory's backing from a single dataset to a backing-agnostic resolver over STAC collections/items (and MosaicJSON). Foundational, deferred work; filed by the roadmap-realignment audit (it threads through Story 9 and the EPIC's "readers (COG, STAC, MosaicJSON)" acceptance criterion, but had no issue).

## Context

Today the resolver is single-dataset (`url=` / `DatasetParams`, 2-D, no `t`). [ADR-0001](https://github.com/developmentseed/titiler-covjson/blob/main/docs/adr/0001-covjson-http-api-direction.md) designed the resolver as a single backing-agnostic seam and **deliberately deferred** a multi-source backing, explicitly rejecting a premature `titiler-stacapi` coupling.

## Why it is foundational

Expanding the seam unlocks three capabilities at once:

- **A real `t` source** (STAC item datetimes) -> the Temporal endpoint surface (the temporal umbrella depends on this).
- **EDR `collectionId` addressing** (STAC collection IDs double as EDR `collectionId`s).
- **Multi-item CoverageCollection** (#24).

It also completes the EPIC's "readers (COG, STAC, MosaicJSON) works" acceptance criterion -- only COG is wired today.

## Options

STAC-first via `titiler-stacapi` is the recommended option; a time-dimensioned NetCDF/Zarr via xarray is an alternative (ADR-0001).

## Status

Foundational / deferred. No milestone. **Blocks** the temporal-surface umbrella (#68) and #24 (CoverageCollection).

Start with docs/adr/0001-covjson-http-api-direction.md and trace the current single-dataset resolver using url= and DatasetParams. Define the scope against the stated STAC-first option and its temporal, collectionId, CoverageCollection, and MosaicJSON requirements; done means the resolver seam supports the agreed multi-source backing without premature coupling.

Written by the indexing model from the issue text.