Support targets file names with different file extension than x file names

[AlexandreBrown]

🚀 Feature

The feature consists in supporting the use case where we have train/val/test images with an extension and train/val/test targets with a different extension.

eg:

train_folder

• File extension : .jpg

train_target_folder

• File extension : .tiff

val_folder

• File extension : .jpg

val_target_folder

• File extension : .tiff

Motivation

This is a common use case that should be supported in order to open Lightning Flash to a broader audience.
This also reduces the barrier to try out Flash by eliminating the need to convert your data to a matching file extension.

Currently this is not supported if your target file paths have a different file extension than the X file paths :

data_module = SemanticSegmentationData.from_folders(
    train_folder=train_images.path,
    train_target_folder=train_masks.path,
    val_folder=valid_images.path,
    val_target_folder=valid_masks.path,
    num_classes=number_of_classes,
    batch_size=batch_size
)

Actual Outcome : ValueError: not enough values to unpack (expected 2, got 0)
Expected Outcome : No error message and it works

Additional context

See slack discussion that led to this issue https://pytorch-lightning.slack.com/archives/CRBLFHY79/p1642626657155300