Review and load an unfamiliar CSV

// Synthetic sensors. Run from the directory containing examples/fixtures.
// Review the recommendation before using an unfamiliar source scientifically.
import aner.data
import aner.units

let proposed = data.infer_csv("examples/fixtures/data_modern.csv",
    missing: "NA", memory_bytes: 8388608, sample_rows: 0)
print(data.schema_description(proposed))
let source = data.scan_csv("examples/fixtures/data_modern.csv", proposed, "NA")

// Fill is a declared source-unit literal; normalization follows its source unit.
// This teaching constant is not a learned imputation estimate.
let filled = data.fill_missing(source, "Temperature_DegC", "21.0")
let normalized = data.convert_unit(filled, "Distance_cm", "m")
let complete = data.drop_missing(normalized)
let report = data.summarize(complete, max_bytes: 8388608, batch_rows: 2)
print("Complete selected rows")
print(data.rows(report))
print("Mean temperature in degC")
print(data.unit_mean(report, "Temperature_DegC", "degC").value())
print("Execution and source identity")
print(data.provenance(report))

// Zero removes the row quota; retained rows automatically spill as RAM fills.
let table = data.collect(complete, max_rows: 0, max_bytes: 8388608, batch_rows: 2)
print(data.text(table, 0, "ObservationID"))
print(data.datetime_ns(table, 0, "RecordedAt"))

// For datasets larger than RAM, stream selected rows to a fresh disk path:
// let saved = data.write_csv(complete, "clean-sensors.csv", 8388608, 4096)
// print(data.provenance(saved))