feat(cli): add export --downsample <seconds> for thinner exports

[jongio]

What

Adds a --downsample <seconds> flag to thermoworks export. It keeps at most one reading per channel per fixed time bucket, so a long cook logged every few seconds can be exported as a per-minute or per-five-minute series.

thermoworks export ABC123 --downsample 60 --format csv --output brisket.csv
thermoworks export ABC123 --downsample 300 --format json

Why

Archive exports include every logged reading, which for a multi-hour cook can be thousands of rows per channel. That is a lot to chart or store when a coarser series would do. Downsampling thins the output without needing a second tool.

How

• --downsample takes a positive integer number of seconds and is off by default (every reading is exported, unchanged behavior).
• A pure downsampleRows(rows, intervalSeconds) helper keeps the earliest reading in each channel's time bucket. Buckets are aligned to the Unix epoch, so the same instants bucket the same way across channels and repeated exports.
• Rows with an unparseable timestamp are always kept, so no data is silently dropped.
• Downsampling runs on the flattened rows before formatting, so it applies to csv, json, and influx output alike.

Testing

• pnpm --dir packages/cli typecheck - clean
• pnpm --dir packages/cli test - 886 passing (10 new)
• pnpm lint - clean

New tests cover parseExportArgs flag validation (positive integer, missing value, non-integer, zero/negative), the downsampleRows helper (bucketing per channel, earliest-in-bucket, unparseable timestamps, empty input), and an end-to-end exportData --downsample run.

Closes #275