[Fix] Read TensorCast's WORLD placement under its current names

[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

When no placement is passed in, the TensorCast host allocator reads get_parallel().world_rank and get_parallel().world_size. Those names had already been split into launch_world_rank / launch_world_size when the backend landed, so both production paths (TensorcastStore.__init__ and get_tensorcast_host_allocator_from_runtime) raise AttributeError. The package's own TensorCast tests miss this: they replace get_parallel with a namespace that carries the old names.

Modifications

• Read launch_world_rank / launch_world_size, and give the in-package test stubs the same names.
• Add a census to test_runtime_context.py: every get_parallel().<name> read in the package, direct or through a name bound to the context, must name a declared parallel field. The one remaining read of an undeclared name, the KV-shard control plane's enable flag (mem_cache/page_interleave.py, currently unreachable), is listed with its reason. The census fails once that read goes away, so the exception cannot go stale.

Accuracy Tests

Not applicable: no model output changes.

• On main, the census reports exactly the two TensorCast reads; on this PR it passes.
• test/registered/unit at this PR's head, compared with main: no new failures.
• Not run: the TensorCast package tests (they need tensorcast installed).

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 #36772466190
Latest PR Test (Extra): 🚫 Run #36772465602
Latest PR Test (AMD ROCm 10): 🚫 Run #36772465942