config: three cache and pool readers take the bags

[ch-wan]

Motivation

pool_configurator, mem_cache/registry and unified_radix_cache read resolved configuration
off a ServerArgs they were handed. The record holds the operator's raw input, so a
resolution-filled field read there answers with the pre-resolution value; the published bag is
what resolution decided.

Modifications

Eight reads move: max_total_tokens and page_size to get_schedule(), enable_hisparse /
radix_cache_backend / hicache_host_memory_mode / enable_session_radix_cache to
get_memory(), enable_streaming_session to get_serving(), extra_metric_labels to
get_observability().

max_speculative_num_draft_tokens stays on the record: it is a derived property, not a field,
so no bag serves it.

test_pool_configurator read page_size off the record too, and now reads the same bag.

These three are the first of eight modules that the launch stand-in's write-through currently
props up. The set was measured, not guessed: deleting
_apply_fields(server_args, self._fields) from override_server_args and running the 159
registered tests that mention the record leaves 27 failures across 11 files, of which 7 are the
hook's own guards. Converting these readers is a step toward retiring that write-through, and
the remaining five modules follow the same shape.

One caveat, learned by trying the wide version of this change first: whether a read can move
to a bag depends on whether the reader's process publishes, not on whether the field maps to a
namespace. A mechanical sweep of all 94 such reads produced ~180 test failures, because many
unit tests hand a record in and never publish, and because HttpServerEngineAdapter
constructs a ServerArgs in the caller's process and never publishes at all. So this goes one
module at a time, with its tests.

Accuracy Tests

2171 tests across mem_cache/, model_executor/ and observability/ pass, and the 159
registered tests that mention the record report the same failure set as the base.

Speed Tests and Profiling

None. A bag read replaces an attribute read.

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

Eight lines of read, one line of test. The question to ask of each is whether the reader runs
after a publish; all three of these are scheduler-side.

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.
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 #33194280480
Latest PR Test (Extra): 🚫 Run #33194280151
Latest PR Test (AMD ROCm 7.2): ❌ Run #33194280257