[Refactor] Stop passing models' layers the placement they already read

[ch-wan]

This PR is part of a stack (oldest at bottom):

• [Feature] Add --attn-dp-size and deprecate --enable-dp-attention #41818
• [Refactor] Enter draft TP scopes by attention ownership and drop ModelRunner.tp_group #41817
• [Refactor] Add make_pp_layers so models stop handing their PP position down #41816
• -> [Refactor] Stop passing models' layers the placement they already read #41815
• [Refactor] Let FusedMoE's weight-loading helpers read the layer's MoE-TP rank #41814
• [Refactor] Keep only the TP and PP groups on the model runner #41813
• [Refactor] Drop model placement attributes and parameters nothing reads #41812
• [Refactor] Drop placement values nothing reads #41811
• [Fix] Make the Solar model constructible and runnable #41810
• [Fix] Size the prefill delayer's gather buffer to its TP group #41809
• [Fix] Pass the draft's attention ownership to DFLASH's eager LiLiCorr scope #41808
• [Fix] Shard MoE WNA16 and Quark INT4-FP8 weights by the MoE placement #41807
• [Fix] State the configured MoE-DP width in the weight-cache fingerprint #41806
• [Fix] Read TensorCast's WORLD placement under its current names #41805

Motivation

Several model modules hand a layer exactly the placement that the layer resolves by itself. That reads as a deliberate choice of group when there is none.

Modifications

• Grok's attention passed tp_rank / tp_size (in locals misnamed attn_tp_*) to its QKV and output projections, whose defaults are those values. Keep only the TP width for the head arithmetic.
• ZAYA's attention passed its stored TP rank and size to CCA and to o_proj, which default to the same values. Its tp_rank copy is then unread and goes too.
• IQuest-Q1 passed kv_tp_rank / kv_tp_size equal to the tp_rank / tp_size of the same call, which is what QKV defaults them to.
• GLM-5 Next gave sharded_weight_loader a getter for the attention-TP rank, which is what the loader reads when no getter is given.
• The DSpark draft attention passed MqaAttentionBase the attention-TP rank and size it defaults to, and no other caller sets them. Drop the two parameters and keep the attributes, which decode-time attention TP still adjusts.

No layer sees a different value.

Accuracy Tests

Not applicable: no layer receives a different value.

• All modified modules import.
• test/registered/unit at this PR's head, compared with main: no new failures.

Speed Tests and Profiling

Not applicable.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): 🚫 Run #36772845494
Latest PR Test (Extra): 🚫 Run #36772845189
Latest PR Test (AMD ROCm 10): 🚫 Run #36772845538