Update the Ironwood offloading description

[RissyRan]

Description

Harini reached out and pointed out those xla_tpu_enable_sparse_core_collective_offload* flags are enabled by default on Ironwood, not like v6e or v5p (I also cross checked in this doc). Reword the document a little bit and recommend customers to tune flags if needed.

# Original:
Enable SparseCore offloading for collectives: By setting the appropriate [XLA flags](https://github.com/AI-Hypercomputer/maxtext/blob/ed517cf80d9aa81f76e236c5516dacebfe39e96d/benchmarks/xla_flags_library.py#L70-L116), you can offload collective operations (like All-Reduce, All-Gather, etc.) to the SparseCores. These operations then run in parallel with the TensorCore computations, effectively hiding communication latency and improving Model Flop Utilization (MFU).

# Updated:
Leverage SparseCore offloading: By default, collective operations (like All-Reduce, All-Gather, etc.) are offloaded to SparseCore, allowing them to run in parallel with TensorCore computations. This effectively hides communication latency and improving Model Flop Utilization (MFU). Also, you can maximize throughput tuning those [XLA flags](https://github.com/AI-Hypercomputer/maxtext/blob/ed517cf80d9aa81f76e236c5516dacebfe39e96d/benchmarks/xla_flags_library.py#L70-L116).

Also format change to pass the mdformat check.

Tests

Manual check

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[harini-sridhar]

LGTM

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[NuojCheng]

Thank you Ran!