perf: cache inspect.signature and UntrackedValue isinstance scan

[John Kennedy (jkennedyvz)]

Summary

• Cache inspect.signature results in RunnableCallable.__init__ via a module-level _SIGNATURE_CACHE dict, avoiding repeated signature introspection for the same function (e.g. 1000 ChannelWrite instances all inspecting the same _write/_awrite methods)
• Cache the any(isinstance(ch, UntrackedValue) ...) scan as _has_untracked_channels once in __enter__/__aenter__, replacing two per-call scan sites in put_writes and checkpoint sanitization — eliminates 1M+ isinstance calls through the ABC machinery for sequential_1000

Split out from #6969 to keep cache changes separate from other optimizations.

Test plan

• Existing test suite passes (make test in libs/langgraph)
• Benchmark confirms no regression

🤖 Generated with Claude Code

[John Kennedy (jkennedyvz)]

From CI

WARNING: the benchma
Notice: +-----------------------------------------+---------+-----------------------+
| Benchmark                               | main    | changes               |
+=========================================+=========+=======================+
| sequential_1000_sync                    | 371 ms  | 233 ms: 1.59x faster  |
+-----------------------------------------+---------+-----------------------+
| sequential_1000                         | 438 ms  | 299 ms: 1.47x faster  |
+-----------------------------------------+---------+-----------------------+
| pydantic_state_9x1200_checkpoint        | 70.6 ms | 63.6 ms: 1.11x faster |
+-----------------------------------------+---------+-----------------------+
| fanout_to_subgraph_10x                  | 33.5 ms | 30.6 ms: 1.10x faster |
+-----------------------------------------+---------+-----------------------+
| fanout_to_subgraph_10x_checkpoint       | 35.2 ms | 32.3 ms: 1.09x faster |
+-----------------------------------------+---------+-----------------------+
| pydantic_state_25x300_checkpoint_sync   | 40.6 ms | 37.6 ms: 1.08x faster |
+-----------------------------------------+---------+-----------------------+
| pydantic_state_25x300                   | 23.8 ms | 22.1 ms: 1.07x faster |
+-----------------------------------------+---------+-----------------------+
| pydantic_state_15x600_checkpoint        | 79.4 ms | 74.2 ms: 1.07x faster |
+-----------------------------------------+---------+-----------------------+
| wide_state_25x300_sync                  | 10.4 ms | 9.77 ms: 1.07x faster |
+-----------------------------------------+---------+-----------------------+
| wide_dict_25x300_sync                   | 10.3 ms | 9.61 ms: 1.07x faster |
+-----------------------------------------+---------+-----------------------+
| pydantic_state_25x300_checkpoint        | 44.8 ms | 42.2 ms: 1.06x faster |
+-----------------------------------------+---------+-----------------------+
| wide_state_15x600_sync                  | 13.5 ms | 12.8 ms: 1.06x faster |
+-----------------------------------------+---------+-----------------------+
| wide_dict_15x600_sync                   | 13.3 ms | 12.6 ms: 1.06x faster |
+-----------------------------------------+---------+-----------------------+
| fanout_to_subgraph_100x_sync            | 297 ms  | 295 ms: 1.01x faster  |
+-----------------------------------------+---------+-----------------------+
| react_agent_10x_sync                    | 22.2 ms | 22.2 ms: 1.00x faster |
+-----------------------------------------+---------+-----------------------+
| fanout_to_subgraph_100x_checkpoint_sync | 313 ms  | 313 ms: 1.00x slower  |
+-----------------------------------------+---------+-----------------------+
| react_agent_10x                         | 26.8 ms | 26.8 ms: 1.00x slower |
+-----------------------------------------+---------+-----------------------+
| serde_allowlist_large                   | 183 us  | 184 us: 1.00x slower  |
+-----------------------------------------+---------+-----------------------+
| wide_state_15x600_checkpoint            | 53.5 ms | 54.0 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| sequential_10                           | 2.38 ms | 2.40 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| sequential_10_sync                      | 1.74 ms | 1.76 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| wide_dict_15x600_checkpoint_sync        | 47.6 ms | 48.1 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| wide_state_15x600_checkpoint_sync       | 47.8 ms | 48.4 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| wide_dict_15x600_checkpoint             | 52.9 ms | 53.6 ms: 1.01x slower |
+-----------------------------------------+---------+-----------------------+
| serde_allowlist_small                   | 39.0 us | 39.6 us: 1.01x slower |
+---------------------------------------