Adjust resource-sensitive MACA examples for contiguous inputs and smaller launch configurations

[VitalyAnkh]

Fixes #29.

Review note

This branch now carries only the shared MACA backend prerequisite from #33 plus the resource-portability changes below. It no longer includes the low-precision example work from #34.

Problem

A separate group of MACA failures is caused by resource assumptions at the example layer. The algorithms are sound, but some examples pass non-contiguous inputs into TileLang kernels or use launch settings that exceed practical MACA limits.

What this PR changes

• materialises contiguous inputs before launching the affected kernels
• reduces launch shapes and related tuning parameters for MACA in the resource-bound examples
• broadens the optional-import fallback in the GDN examples so the intended TileLang path remains reachable
• removes or reduces CUDA-oriented shared-memory layout assumptions where MACA has tighter limits

Solution

The fix is to make the examples portable with respect to layout and resource usage. By ensuring contiguity and choosing MACA-safe launch parameters, the same kernels can execute correctly without inflating the example logic with hidden assumptions about a larger CUDA/Hopper resource budget.

Alternatives considered

One alternative was to leave the examples unchanged and attempt to compensate in backend scheduling. That would have obscured the true issue, namely that these examples had been tuned for a larger CUDA/Hopper resource budget. Another option was to skip the offending cases on MACA, but that would have left straightforward portability work unresolved.

Verification

• python -m pytest -q examples/maca/attention_sink/test_example_attention_sink.py::test_example_gqa_sink_varlen examples/maca/convolution/test_example_convolution.py::test_example_convolution examples/maca/deepseek_mla/test_example_mla_decode.py::test_example_mla_decode examples/maca/flash_attention/test_example_flash_attention.py::test_example_gqa_bwd_tma_reduce_varlen examples/maca/gdn/test_example_gdn_compilation.py

[github-actions]

👋 Hi! Thank you for contributing to the TileLang project.

Please remember to run pre-commit run --all-files in the root directory of the project to ensure your changes are properly linted and formatted. This will help ensure your contribution passes the format check.

We appreciate you taking this step! Our team will review your contribution, and we look forward to your awesome work! 🚀