feat: support weight upload for yolo26-sem semantic segmentation models

[leeclemnet]

Description

Prior behavior: The SDK had no awareness of semantic segmentation as a
distinct task. TYPE_SEMANTIC_SEGMENTATION projects were explicitly skipped
in upload validation, and SemanticSegmentationModel checkpoints (new in
ultralytics 8.4.55) were unrecognised.

New behavior: A new TASK_SEM = "sem" constant is wired through the
weight upload pipeline:

• task_of_model_type("yolo26-sem") → TASK_SEM
• _detect_yolo_task() recognises SemanticSegmentationModel checkpoints
• validate_model_type_for_project() enforces that -sem models match
  semantic-segmentation projects (and rejects mismatches in both directions)
• Auto-suffix: declaring plain "yolo26" with a semantic-seg checkpoint
  auto-corrects to "yolo26-sem"

No changes needed to supported_models, dependency checks, or artifact
extraction — "yolo26" in "yolo26-sem" passes all existing gates.

Motivation: roboflow-train and roboflow-model-conversion already support
yolo26-sem; the SDK was the missing link for user-side weight upload.

Type of change

• New feature (non-breaking change which adds functionality)

How has this change been tested, please provide a testcase or example of how you tested the change?

• unit tests
• versionless weight upload (pre-trained cityscales yolo26n-sem) succeeds on prod

- [x] versioned weight upload (same model) succeeds on staging with https://github.com/roboflow/roboflow/pull/12297

Will the change affect Universe? If so was this change tested in universe?

n/a

Github Actions Required to Deploy

n/a — SDK package, no deployment needed.

Infrastructure impact

• This change affects infrastructure needs (GPUs, Cloud Function sizing, storage etc.)
• This change requires updates to secrets
• This change materially impacts cloud costs (explain briefly)

Docs

• This change requires a documentation update
• Docs updated? What were the changes: