S3-ocr

- Pages with no OCR text on them are now recorded as rows with empty strings, instead of being skipped entirely. [23](https://github.com/simonw/s3-ocr/issues/23)

- Fixed bug where commands were sometimes not properly registered. [26](https://github.com/simonw/s3-ocr/issues/26)

- Now pins to `click>=8.0`, which should avoid a bug where installing this on a machine with an older version of Click present would lead to the commands failing to register. [25](https://github.com/simonw/s3-ocr/issues/25)
- `s3-ocr --help` now includes links to the documentation and changelog.

- `s3-ocr start` now automatically pauses and then retries if Textract complains that there are too many jobs running. This can be turned into an early exit with an error message using the new `--no-retry` option. [21](https://github.com/simonw/s3-ocr/issues/21)
- New `s3-ocr start --dry-run` option for displaying what would happen without starting the OCR process. [22](https://github.com/simonw/s3-ocr/issues/22)
- Textract now runs in the same region as the S3 bucket it is writing to, avoiding an error. [24](https://github.com/simonw/s3-ocr/issues/24)

- Ability to run OCR against just the PDF files contained within a specific folder in the S3 bucket, using `s3-ocr start my-bucket --prefix my-prefix/`. [20](https://github.com/simonw/s3-ocr/issues/20)
- New command: `s3-ocr dedupe my-bucket` - scans the bucket for any new files that are duplicates of files that have already been OCRd and writes out job results to reuse existing OCR results and avoid processing them a second time in the future. [19](https://github.com/simonw/s3-ocr/issues/19)

- New command: `s3-ocr inspect-job <job_id>` returns information about the status of a specific job. [15](https://github.com/simonw/s3-ocr/issues/15)
- Added a live demo at [s3-ocr-demo.datasette.io](https://s3-ocr-demo.datasette.io/). [#16](https://github.com/simonw/s3-ocr/issues/16)