[REFACTOR][IR] attrs.h follow-up cleanup: drop legacy vtable / rename / phase out AttrFieldInfo

[tqchen]

Summary

Follow-up to #19607 that continues trimming attrs.h and adjacent
files. The six commits land independently and each builds clean.

• Phase out OpNode::arguments and AttrFieldInfo — the field stored
  metadata that no Python tooling, test, or C++ caller (beyond internal
  sanity checks) read; removing it deletes AttrFieldInfo plus ~335
  chained .add_argument(...) calls. The remaining 12 internal consumers
  now read op->num_inputs and report indexed inputs (input[i]).
• Drop the (unused) virtual destructor on BaseAttrsNode (ffi::Object
  uses a captured-typed deleter, no virtual dispatch needed) and inline
  the trivial 3-line DictAttrs(Map) constructor into the header.
• Rename BaseAttrsNode → AttrsNode; the Base prefix existed only
  to distinguish from the AttrsNodeReflAdapter shim that [REFACTOR][IR] Cleanup attrs.h: drop NullValue, AttrsNodeReflAdapter, legacy BaseAttrsNode methods #19607
  removed. The "ir.Attrs" FFI registry key is unchanged.
• Promote DictAttrs to NOTNULLABLE
  (TVM_FFI_DEFINE_OBJECT_REF_METHODS_NOTNULLABLE + COW macro). The
  no-arg DictAttrs() constructor already created an empty backing,
  so every existing call site already produced a defined object;
  ~15 defensive attrs.defined() checks (and a defensive Python None
  fallback in Function) are now redundant.
• Inline the WithAttr(DictAttrs, ...) / WithAttrs(DictAttrs, ...)
  free-function overloads into the TFunc-template wrappers — those
  overloads had no external callers (no TVM_DLL, no Python binding).
• Rename AttrsWithDefaultValues<T> → PassConfigWithDefaults<T> and
  move from attrs.h to transform.h; all 9 consumers are pass-config
  classes registered via TVM_REGISTER_PASS_CONFIG_OPTION.

attrs.h shrinks from 363 → 262 lines.

[gemini-code-assist]

Code Review

This pull request refactors the attributes system in TVM, renaming BaseAttrsNode to AttrsNode, simplifying DictAttrs and Attrs definitions, and removing the add_argument method and argument descriptions from operator registrations. It also updates several passes to use PassConfigWithDefaults. The reviewer feedback strongly cautions against the widespread removal of defined() checks on attributes across various files (including attrs.h, expression constructors, printers, and transform passes). Since DictAttrs can still be null or undefined at runtime, removing these checks introduces a high risk of segmentation faults when accessing attrs->dict, and the reviewer recommends restoring these safety checks.

[gemini-code-assist]

If node->attrs is null (which can happen if the function was deserialized from a format where attributes were omitted), calling CopyOnWrite() on it will cause a segmentation fault. We should defensively initialize node->attrs if it is not defined.

  if (!node->attrs.defined()) {
    node->attrs = DictAttrs();
  }
  node->attrs.CopyOnWrite()->dict.Set(attr_key, std::move(attr_value));

[gemini-code-assist]

If node->attrs is null, calling CopyOnWrite() on it will cause a segmentation fault. We should defensively initialize node->attrs if it is not defined.

Suggested change

	auto* dict_node = node->attrs.CopyOnWrite();
	if (!node->attrs.defined()) {
	node->attrs = DictAttrs();
	}
	auto* dict_node = node->attrs.CopyOnWrite();

[gemini-code-assist]

If node->attrs is null, calling CopyOnWrite() on it will cause a segmentation fault. We should check if node->attrs is defined before attempting to erase from it.

  if (node->attrs.defined()) {
    node->attrs.CopyOnWrite()->dict.erase(attr_key);
  }

[gemini-code-assist]

If op->attrs is null (which can happen if the function was deserialized without attributes or created via legacy APIs), dereferencing it via op->attrs->dict will cause a segmentation fault. It is safer to keep the defined() check.

Suggested change

	if (matches && op->attrs->dict.count(kv.first)) {
	if (matches && op->attrs.defined() && op->attrs->dict.count(kv.first)) {

[gemini-code-assist]

If func->attrs is null, calling func->attrs->dict will cause a segmentation fault. We should check if func->attrs is defined before accessing its dictionary.

Suggested change

	if (!func->attrs->dict.count(tvm::attr::kGlobalSymbol)) {
	if (!func->attrs.defined() || !func->attrs->dict.count(tvm::attr::kGlobalSymbol)) {

[gemini-code-assist]

If cur_func_->attrs is null, calling cur_func_->attrs->dict will cause a segmentation fault. We should check if cur_func_->attrs is defined before accessing its dictionary.

Suggested change

	if (cur_func_->attrs->dict.count(tvm::attr::kSTir)) {
	if (cur_func_->attrs.defined() && cur_func_->attrs->dict.count(tvm::attr::kSTir)) {

[gemini-code-assist]

If func->attrs is null, removing this check will cause a segmentation fault when accessing func->attrs->dict later in the lambda. We should keep the defined() check.

      if (!func->attrs.defined()) {
        return DictAttrs();
      }

      ffi::Map<ffi::String, ffi::Any> dict;

[gemini-code-assist]

If attrs is passed as an undefined DictAttrs (which can happen if a caller passes a default-constructed or null DictAttrs), leaving it uninitialized will result in a null attrs field in the Function node, leading to downstream segmentation faults. We should keep the fallback initialization.

Suggested change

	// Set the function type.
	if (!attrs.defined()) {
	attrs = DictAttrs();
	}
	// Set the function type.

[gemini-code-assist]

If attrs is passed as an undefined DictAttrs, leaving it uninitialized will result in a null attrs field in the PrimFunc node, leading to downstream segmentation faults. We should keep the fallback initialization.

  if (!attrs.defined()) {
    attrs = DictAttrs();
  }

  if (!ret_type.defined()) {