feat: add trackio.sweep() for local grid/random hyperparameter search

[akkiyolo]

Partially addresses #657.

Adds the backend piece of hyperparameter sweep support: trackio.sweep()
runs a local grid or random search over a defined parameter space by
calling a user-supplied training function once per configuration.

• trackio.sweep(sweep_config, function, count=None, sweep_id=None) ,API
  shape roughly mirrors wandb's sweep config (method, parameters with
  values or min/max) so existing configs mostly drop in
• Any trackio.init() called inside function automatically merges the
  sweep-selected hyperparameters into config and groups the run under a
  shared sweep id (unless group is explicitly passed), reusing the
  existing config/group fields ,no schema changes
• Added examples/hyperparameter-sweep.py and tests/unit/test_sweep.py
  (16 tests covering config generation edge cases and init() integration)

Not included in this PR (see plans/issue-657-hyperparameter-sweeps-plan.md
for a scoped breakdown and follow-up ideas):

• Parallel-coordinates chart in the dashboard
• CLI-driven sweeps over an external script
• Bayesian/other advanced search strategies

Ran pytest tests/unit/test_sweep.py (16 passed) and the full unit suite
locally with no new failures, plus ruff check --fix --select I && ruff format.

Commented on the issue beforehand to flag this scope: #657 (comment)