Stop sharing C/C++ code with LiteRT (except schema.fbs)

[veblush]

TFLM currently shares several headers and source files with LiteRT (tensorflow/lite/... and tensorflow/compiler/mlir/lite/...), relying on #ifdef TF_LITE_STATIC_MEMORY to strip mobile & desktop-only code and struct fields. This causes ODR/header collisions when linking both runtimes in the same binary and forces both projects to maintain TF_LITE_STATIC_MEMORY.

We are decoupling TFLM into a completely standalone codebase under tensorflow/lite/micro/ that shares only schema.fbs (tensorflow/lite/schema/schema.fbs) with LiteRT:

• Step 1 (Mostly standalone TFLM #3671): Move tensorflow/lite/kernels/internal/** and C++ kernel helpers (RuntimeShape, reference_ops, quantization_util, kernel_util) into tensorflow/lite/micro/ under namespace tflite::micro (with namespace tflite compatibility aliases).
• Step 2 (Decouple C API from LiteRT #3788): Move C API headers and sources (common.h, common.cc, c_api_types.h, builtin_op_data.h) into tensorflow/lite/micro/c/ while temporarily keeping TF_LITE_STATIC_MEMORY and forwarding headers in tensorflow/lite/{core/,}c/ for downstream migration.
• Step 3 (Remove tensorflow/compiler/ and tensorflow/lite/{c,core,kernels}):
  • Move schema.fbs to tensorflow/lite/schema/schema.fbs and schema_utils.{h,cc} into tensorflow/lite/micro/.
  • Move flatbuffer_conversions.{h,cc} into tensorflow/lite/micro/tflite_bridge/ under namespace tflite::micro and replace TF_LITE_REPORT_ERROR with MicroPrintf.
  • Merge macros.h into tensorflow/lite/micro/kernels/internal/compatibility_macros.h.
  • Delete unused LiteRT files (error_reporter.{h,cc}, micro_error_reporter.{h,cc}, tensor_utils.{h,cc}, context_util.h, tensorflow/lite/array.{h,cc}), forwarding headers (tensorflow/lite/{core/,}c/, tensorflow/lite/kernels/), tensorflow/compiler/, and ci/sync_from_upstream_tf.sh.
• Step 4 (Remove heap-allocating array.{h,cc} helpers):
  • Replace test-only IntArrayUniquePtr usage in micro_allocator_test.cc with IntArrayFromInts, delete tensorflow/lite/micro/array.{h,cc}, and remove TfLiteIntArrayCreate/Free and TfLiteFloatArrayCreate/Free from tensorflow/lite/micro/c/common.{h,cc}.
• Step 5 (Namespace tensorflow/lite/micro/c/ & remove TF_LITE_STATIC_MEMORY):
  • Give tensorflow/lite/micro/c/*.h dedicated TENSORFLOW_LITE_MICRO_C_*_H_ header guards and move C API types/functions into namespace tflite::micro (with global-namespace aliases when LiteRT's common.h is not included).
  • Remove #ifndef TF_LITE_STATIC_MEMORY blocks, -DTF_LITE_STATIC_MEMORY build flags, and test_x86_no_tflite_static_memory.sh.