Make Discovery accessible from CLI or SDK

[kaleko]

Is your feature request related to a problem? Please describe.
If a user has a large number of classes of documents, they would like a programmatic way to run discovery on each without having to upload to the UI and manually download the generated config each time.

Describe the solution you'd like
I'd like to be able to run Discovery with idp-cli or the sdk, so I can do it programmatically.

Describe alternatives you've considered
I've implemented a standalone Discovery module which copies the IDP source code and prompts etc outside of the IDP solution and hits Bedrock directly. Would much rather use IDP directly.

Additional context
N/A

[rstrahan]

This has been implemented in the feature/discovery-cli-sdk branch.

CLI Usage

# Local mode — no deployed stack needed
idp-cli discover -d ./invoice.pdf
idp-cli discover -d ./invoice.pdf --output schema.json

# Stack mode — saves schema to DynamoDB config
idp-cli discover --stack-name my-stack -d ./w2.pdf
idp-cli discover --stack-name my-stack -d ./w2.pdf --ground-truth ./w2-gt.json

# Batch — multiple documents
idp-cli discover --stack-name my-stack -d ./invoice.pdf -d ./w2.pdf -d ./paystub.png

SDK Usage

from idp_sdk import IDPClient

# Local mode (no stack)
client = IDPClient()
result = client.discovery.run("./invoice.pdf")
print(result.json_schema)

# Stack mode (saves to config)
client = IDPClient(stack_name="my-stack")
result = client.discovery.run("./invoice.pdf", config_version="v2")

# Batch
batch = client.discovery.run_batch(["./doc1.pdf", "./doc2.pdf"])

Key design decisions

• No S3 uploads — document bytes read locally, passed directly to Bedrock
• --stack-name optional — local mode uses system default Bedrock settings
• Single discover command — pass -d multiple times for batch
• Ground truth support — --ground-truth for more accurate schema generation
• Backward compatible — ClassesDiscovery new params are all optional

PR: #232

[rstrahan]

Fixed in v0.5.2