Add end-to-end BatchNorm operator tutorial (Chinese)

[Copilot]

Ticket

Link to Github Issue

Problem description

The existing "Adding an Op" tutorial (adding-an-op.md) covers the basics using matmul but lacks depth on normalization ops with multiple inputs, optional operands, multi-output variants, and decomposition patterns. A detailed tutorial using BatchNorm as the reference op was requested.

What's changed

Added docs/src/adding-an-op-batchnorm-tutorial.md — a comprehensive Chinese-language tutorial that walks through all ~25 files touched when adding BatchNorm end-to-end. Uses real code extracted from the existing BatchNorm implementation.

Covers 12 steps across the full compiler pipeline:

• TTIR/TTNN dialect definitions — tablegen, verifiers, comparison of required vs optional operands (AttrSizedOperandSegments)
• Conversion passes — StableHLO→TTIR, TTIR→TTNN (including multi-output handling for Training variant), TTNN→EmitC/EmitPy (prefix swap patterns, keyword arg ordering)
• Flatbuffer schema & serialization — table definitions, OpType union, createOp dispatch
• Runtime — header/impl, CMake registration, program_executor dispatch, input/output refs
• OpModel interface — full implementation and stub patterns
• TTIRBuilder & golden functions — @tag/@parse/@split methods, GoldenMapTensor limitations
• Tests — lit tests, EmitC/EmitPy tests, Python golden tests, StableHLO conversion tests
• TTIR decomposition — NCHW normalization strategy unique to BatchNorm

Also highlights key pitfalls: emitOpError() vs notifyMatchFailure() in pattern rewriters, attribute type mapping (I32Attr → get_signless), EmitPy positional/keyword arg ordering.

Updated docs/src/SUMMARY.md to add navigation entry under "Adding an op".

Checklist

• New/Existing tests provide coverage for changes