Spreadsheet analysis · turn dirty data into conclusions

I exported a full-year 2025 order detail CSV from the order system, 3695 rows in total. I won't send you the data (it contains customer info and is too large),
I'll only give you the header and the first 20 rows of sample data. Based on that, write a Python script that runs directly.
Sample (including header): [paste header + first 20 rows]
The script should do four things and print the result of each:
1. Data profiling: check item by item and report affected row counts (column name conventions, date format variants, non-numeric amount notations,
    missing-value representations, fully duplicate rows, negative or extreme outlier amounts, blanks in key fields).
    The sample is only 20 rows; the real data may be dirty in more ways, so the checks must be robust. For unseen formats, bucket them as "unparseable" and count.
2. Cleaning: clean according to the rules you judge and print the rules. At least handle: unify dates to YYYY-MM-DD, convert amounts to float,
    unify missing values, deduplicate, normalize province names, handle outliers.
3. Statistics: row counts before and after cleaning, full-year net sales, monthly order counts and sales, order counts / sales / refund rate by channel,
    sales by province, max single-customer single-day order count and the top 3.
4. Suspicious items: automatically flag months with abnormal volume, channels with abnormal refund rates, customers with abnormal ordering behavior. Judgment should be based on computed numbers, do not hardcode thresholds.
Hard requirements: use only the standard library (csv/re/collections/datetime/statistics), no pandas;
file path via command-line argument; the file is UTF-8 with BOM so use utf-8-sig; the code must run directly with no placeholders or TODOs; output in Chinese.
Only output one complete Python script, no explanation.