fix: support custom output quantiles via ForecastConfig (#286)

[aydinomer00]

Problem

When passing custom quantiles (e.g. quantiles=[0.3, 0.4]) to ForecastConfig,
the model raised a shape mismatch error because the output architecture was assumed
to always return all 10 default quantile values.

Closes #286

Solution

The model always outputs 10 values internally (1 point + 9 quantiles: 0.1–0.9).
This PR adds an output_quantiles field to ForecastConfig that slices the output
tensor after decoding — no model weights or checkpoint formats changed.

Fully backwards-compatible: output_quantiles=None (default) returns all 9
quantiles as before.

Changes

• src/timesfm/configs.py: Added DEFAULT_QUANTILES constant and
  output_quantiles: tuple[float, ...] | None = None to ForecastConfig
• src/timesfm/timesfm_2p5/timesfm_2p5_torch.py: Validate at compile() and slice output
• src/timesfm/timesfm_2p5/timesfm_2p5_flax.py: Same fix for Flax backend

Usage

model.compile(
    timesfm.ForecastConfig(
        max_context=1024,
        max_horizon=256,
        output_quantiles=(0.1, 0.5, 0.9),
    )
)
point, quantiles = model.forecast(horizon=12, inputs=[...])
# quantiles.shape → (1, 12, 3)  ✓

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.