feat: annotation-overwrite parameter on upload_dataset and the CLI

[tonylampada]

Description

The zip upload API now takes annotationOverwrite (roboflow/roboflow#14928). This sends it.

Default False, so a re-uploaded image keeps its annotations.
Classification defaults to True, which is what that API accepts.
Overridable with upload_dataset(annotation_overwrite=...) and roboflow image upload --annotation-overwrite/--no-annotation-overwrite.

Under review: the parameter is a three-state Optional[bool], where None reads project.type at the call site. Passing False on a classification project reaches the API and returns its 400, rather than being corrected here.

Type of change

• New feature (non-breaking)

How has this change been tested?

• New automated tests added.

Any specific deployment considerations

Client-only. Needs roboflow/roboflow#14928 deployed.

Pipeline

no-mistakes run 01M1CSZWMEXBFX79S3KR932WRA on bc/CMD-27 @ 864163e8 — outcome: checks-passed.

step	status	findings
intent	completed	0
rebase	completed	0
review	completed	2 — dropped-body-test (warning, ask-user): approved by card owner, deletion stands; flag-ignored-per-image (info, no-op)
test	completed	0
document	completed	1 — product-docs-flag (info, ask-user): approved; external roboflow-product-docs entry carried by card owner
lint	completed	0 (ruff + mypy clean)
push	completed	0
pr	completed	0
ci	completed	all checks green

Note: the pipeline's own PR-body update failed on the gh pr edit projects-classic GraphQL deprecation; this block is reconstructed from no-mistakes axi status run evidence.