docs: fix tree packing algorithm name and token alignment constraint

[guozhihao-224]

Summary

Two corrections to the tree training reference documentation:

1. Packing algorithm: Change "first-fit decreasing" to "first-fit". The code in _greedy_build_tries() iterates sequences in original order without sorting by decreasing size. (A proper FFD implementation exists in areal/utils/seqpack.py for trajectory redistribution, but tree packing intentionally uses plain first-fit to preserve prefix locality.)

2. Token alignment constraint: Document the full constraint lcm(BLOCK_SIZE, parallel_size) instead of just BLOCK_SIZE (128). In most configurations where parallel_size divides 128 this equals 128, but the actual code at tree.py:341 uses math.lcm() which can be stricter with unusual parallel configs.

Verification

• Confirmed no sort step in _greedy_build_tries() at tree.py:476-523
• Confirmed math.lcm(BLOCK_SIZE, parallel_size) at tree.py:341
• Git history shows no sort step was ever present or removed

Test plan

• pre-commit run --all-files passes
• ./docs/build_all.sh builds without errors

Ref: #1165 (Findings #6, #7)

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the tree training documentation to clarify that max_tokens_per_mb must be a multiple of the least common multiple of the block size and parallel size. It also updates the description of the greedy packing strategy from first-fit decreasing to first-fit. Feedback was provided to generalize the definition of parallel_size to ensure accuracy across different backend engines like Megatron and Archon.

[gemini-code-assist]

The definition of parallel_size as tp_size × sp_size is accurate for the FSDP backend, but the Megatron and Archon engines currently only pass the tensor parallel degree (tp_size) to the packing function. To ensure the documentation is accurate across all supported backends, consider clarifying that parallel_size represents the product of parallelism dimensions that require block alignment for the specific engine being used.

Suggested change

	`parallel_size = tp_size × sp_size`. In most configurations where `parallel_size`
	`parallel_size` (the product of parallelism dimensions requiring block alignment, e.g., `tp_size × sp_size` for FSDP). In most configurations where `parallel_size`