Fix MeanAveragePrecision on non-contiguous inputs

[mottopanikeiku]

Description:

Use reshape instead of view when flattening binary inputs and multiclass targets, matching the other paths in _prepare_output. This fixes evaluation on valid transposed segmentation tensors. Add four layout regression cases and document the supported layout.

Reproducer:

import torch
from ignite.metrics import MeanAveragePrecision

p = torch.tensor([[[0.9, 0.2], [0.1, 0.8]]]).transpose(1, 2)
y = torch.tensor([[[1, 0], [0, 1]]]).transpose(1, 2)
m = MeanAveragePrecision()
m.update((p, y))
print(m.compute())

Before	After
RuntimeError: view size is not compatible with input tensor's size and stride	1.0

Tests: pytest tests/ignite/metrics/test_mean_average_precision.py -m 'not distributed and not tpu': 15 passed, 36 deselected. Three new cases fail before the fix; the fourth is an existing-supported-layout control. Ruff check and format check pass.

Check list:

• New tests are added (if a new feature is added)
• New doc strings: description and/or example code are in RST format
• Documentation is updated (if required)

I used AI assistance (Claude and GPT-6 Astra) to find, fix and test this bug.