Tool callbacks skipped when input fails schema validation: no handleToolStart, no handleToolError

Finding: In @langchain/core 1.2.17 and main (commit aa1b519), as reported, invoking a structured tool with callbacks and an input that fails schema validation throws ToolInputParsingException without calling handleToolStart or handleToolError. Valid input fires handleToolStart and handleToolEnd normally (https://github.com/langchain-ai/langchainjs/issues/11887). Reported environment: @langchain/core 1.2.17, langchain 1.5.15, zod 4.6.5, Node.js 22.22 on macOS. The reporter includes a runnable example with a zod schema {a: number, b: number} and a callback handler. Calling add.invoke({a:'one', b:2}) produced only the thrown ToolInputParsingException in the reporter's output. The reporter says Zod and JSON schemas behave the same. Impact described by the reporter: in agents, when a model sends wrongly typed arguments, the failed tool run does not appear in LangSmith. The trace shows the model's tool call followed by the next model turn with no failed tool run in between. Callback handlers that count tool errors also miss these failures. Reported cause (not maintainer-confirmed): StructuredTool.call validates input before calling handleToolStart, so the exception is thrown before the tool run exists. The reporter contrasts this with Python's BaseTool.run, which calls on_tool_start first and validates inside the try block, so validation errors reach on_tool_error. Proposed fix status: the reporter suggests starting the run before validation and passing the exception to handleToolError before rethrowing, leaving the error and ToolMessage unchanged. They say they tested this on 1.2.17 and are opening a PR. No PR link, merge or release was supplied, so the fix is unverified and unreleased as far as the sources show. Limitations: this is a single external report. KFA did not reproduce it. No maintainer response or comments were available. Treat the cause analysis as the reporter's hypothesis. Until a fix is confirmed, validation-failure tool calls may be invisible to callback-based tracing and error counting in affected versions.