Handle Azure Date column inferred as DateType by Spark 4

[jnioche]

Fixes #246

Spark 4 infers ISO-formatted Date columns in Azure CSV exports as DateType, so the getString() call in the azure FOCUSColumns module threw a ClassCastException. As discussed on the issue, this keeps the richer type rather than casting back to string.

Changes

• SparkJob: the Azure CSV post-load fix-ups are extracted into a testable normalizeAzureColumns() method, which now also normalises Date to DateType with coalesce(try_to_date(...)) over the known formats (ISO, MM/dd/yyyy, M/d/yyyy). try_to_date rather than a cast because ANSI mode makes a failed cast throw instead of yielding null.
• AzureColumn: DATE is now declared DateType.
• RowColumn: new getDate(Row) accessor returning LocalDate; accepts java.sql.Date, LocalDate (Java 8 datetime API) and falls back to parsing strings.
• azure/FOCUSColumns: reads the date with getDate() and does plusDays(1); the format-guessing nextDay() is gone. Side effect: ChargePeriodStart/End are now always emitted as ISO dates, even for MM/dd/yyyy inputs.
• pom.xml: surefire gets the standard Spark-on-JDK17+ --add-opens list (keeping the jacoco agent via @{argLine}), needed for the new test.
• Tests: FOCUSColumnsTest builds rows with real date values plus string-fallback coverage, and a new SparkJobTest reads two small CSVs through actual inferSchema — asserting the inference premise and that enrichment no longer throws — since rows built by hand could not catch this.

Verification

Beyond the unit tests (316 pass), ran SparkJob end-to-end on azure-examples/EA-Cost-Actual.csv both as-is (US dates) and with the Date column converted to ISO (the Spark 4 DateType inference scenario from the issue): both runs complete, Date lands as a proper DATE in the output parquet, charge periods are ISO strings and the emissions columns are populated.

Left out deliberately

Typing ChargePeriodStart/End as dates (raised in the issue discussion): the AWS bridge writes hourly CUR timestamps into the same columns, so that needs its own PR.