Replace nested static_for lambdas with compile-time search helper

[tenpercent]

Summary

• Add find_in_tuple_of_sequences compile-time search helper with O(1) template depth
• Replace nested static_for lambdas in TensorDescriptor::GetTransformAndItsUpperDimension
• Replace generate_tuple lambda in TensorDescriptor::InitializeElementSize with pack expansion
• Apply same optimizations to TensorAdaptor

Motivation

The TensorDescriptor and TensorAdaptor classes had excessive template instantiation from:

1. Nested static_for loops with lambdas (918 applier::operator() instantiations)
2. generate_tuple with lambdas (78+ instantiations per class)

Results (example_grouped_conv_fwd_xdl_fp16)

Metric	Before	After	Improvement
Template instantiation time	23.4s	19.1s	18% reduction
applier instantiations	1132	127	89% reduction
generate_tuple lambdas	178	96	46% reduction

find_in_tuple_of_sequences Helper

Uses O(1) template depth via pack expansion instead of O(N) recursion:

Metric	Recursive	O(1) Pack	Improvement
Instantiations	541	273	50% reduction
Time	430ms	133ms	69% reduction

Test Plan

• Waiting for full CI

PR Stack

#	PR	Description
1	#3585	sequence_gen with __make_integer_seq
2	#3588	generate_identity_sequences helper
3	#3589	Named functors in transform_tensor_descriptor
4	#3590	container_concat optimization
5	#3596	O(1) pack expansion rewrites
6	#3600	TensorDescriptor/TensorAdaptor lambda elimination