[REFACTOR][NODE] Use fn_repr inside kRepr lambdas, not ffi::ReprPrint

[tqchen]

Inside a kRepr lambda, fn_repr is the canonical dispatch. Threading it through to sub-element dispatch preserves recursion context and matches the repr migration's design intent (where fn_repr is the core abstraction for repr delegation). This applies the recommendation from code review of #19461 to three sites.

• src/relax/ir/dataflow_pattern.cc — helper threads fn_repr
• src/s_tir/meta_schedule/arg_info.cc — lambda parameter named
• src/target/virtual_device.cc — fn_repr(target) replaces target->str()

[gemini-code-assist]

Code Review

This pull request replaces the legacy ReprPrinter with ffi::ReprPrint, involving the deprecation of the old header and widespread updates to object representation dispatches. The review feedback highlights critical omissions where custom representations for TraceNode and several Py* nodes were removed without being migrated to the new kRepr attribute registration, potentially causing regressions in debug output. Additionally, a static method call in a documentation example requires class qualification for correctness.

I am having trouble creating individual review comments. Click here to see my feedback.

src/s_tir/schedule/trace.cc (549-553)

The ReprPrinter dispatch for TraceNode is removed here, but no corresponding kRepr registration is added to replace it. This will cause Trace objects to lose their custom Python-like representation and fall back to a generic object dump. Since TraceAsPythonRepr is still defined in this file, you should register it as the kRepr attribute for TraceNode within a TVM_FFI_STATIC_INIT_BLOCK.

include/tvm/node/functor.h (186)

The Script method is a static member of TVMScriptPrinter. Since the lambda is defined outside the class scope in this example, it needs to be qualified with the class name to compile correctly.

 *    return TVMScriptPrinter::Script(n->a, cfg) + " + " + TVMScriptPrinter::Script(n->b, cfg);

src/s_tir/meta_schedule/cost_model/cost_model.cc (66-73)

Removing the custom ReprPrinter dispatch for PyCostModelNode without registering a kRepr attribute that calls f_as_string will result in a regression where the Python-provided string representation is lost. The default reflection-based repr will only show the fields of the C++ wrapper (like the f_as_string PackedFunc pointer). This pattern appears in several other Py* nodes in this PR and should be addressed to maintain debugability.