Add real-world examples: data pipeline, MCP search, and naive vs compiled comparison

[Copilot]

The repo had only one trivial arithmetic example, leaving users without concrete evidence of real-world applicability or the latency/cost argument for adopting ChainWeaver.

New examples

• examples/etl_flow.py — 5-step ETL flow (fetch_data → validate_records → normalize_fields → enrich_records → store_records) over an inventory mock dataset. Demonstrates field normalization, type coercion, and invalid-record dropping.

• examples/mcp_search_flow.py — MCP-style retrieval flow (search_knowledge_base → extract_relevant_fields → format_response) over a mock knowledge base. Illustrates how ChainWeaver eliminates intermediate LLM calls from fully deterministic search/extract/format patterns.

• examples/naive_vs_compiled.py — Side-by-side timing comparison of a 5-step record-enrichment flow. The naive path inserts time.sleep(0.3) between each step (simulated LLM round-trips); the compiled path runs the same flow via a registered Flow. Prints a table:

  ┌─────────────────────────┬───────────────┬───────────────┐
  │ Metric                 │        Naive │     Compiled │
  ├─────────────────────────┼───────────────┼───────────────┤
  │ Wall time (ms)         │       1200.7 │          0.3 │
  │ LLM calls              │            4 │            0 │
  │ Speedup                │         1.0x │      4447.3x │
  │ Time saved (ms)        │              │       1200.4 │
  │ LLM calls saved        │              │            4 │
  └─────────────────────────┴───────────────┴───────────────┘
  

README

Updated Quick Start and Development sections to list all four examples with one-line descriptions.

Notes

• The original issue referenced examples/data_pipeline_flow.py; the file was delivered as examples/etl_flow.py — a more descriptive name that avoids the forbidden "pipeline" vocabulary.
• Pre-existing test failure (TestToolZeroDivisionError::test_zero_division_error_wrapped) is not introduced by this PR. Tracked in Fix TestToolZeroDivisionError assertion to match wrapped executor error message #120.

[Copilot]

Pull request overview

Adds multiple runnable, real-world-focused example scripts to demonstrate ChainWeaver’s deterministic execution benefits (including a latency/cost comparison), and updates the README to reference them.

Changes:

• Added three new example scripts: ETL-style data flow, MCP-style search flow, and naive-vs-compiled timing comparison.
• Updated README Quick Start and Development sections to list all available examples and how to run them.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 9 comments.

File	Description
README.md	Lists additional example scripts in Quick Start and Development sections.
examples/data_pipeline_flow.py	New 5-step ETL-style deterministic flow example over mock inventory data.
examples/mcp_search_flow.py	New 3-step MCP-style search → extract → format deterministic flow example.
examples/naive_vs_compiled.py	New timing comparison example contrasting simulated naive LLM step transitions vs FlowExecutor execution.