Allow non-aggregated memray captures

# Feature Request

## Is Your Feature Request Related to a Problem? Please Describe.

Currently, `pytest-memray` generates aggregated capture `.bin` files by default [#107](https://github.com/bloomberg/pytest-memray/issues/107) [Release 1.6.0](https://github.com/bloomberg/pytest-memray/releases/tag/1.6.0).
While this is efficient for disk usage, aggregated captures cannot be used with several memray features (e.g. `memray stats` or other reporters that require per-allocation data). Attempting to use these features results in errors such as:

```
NotImplementedError: Can't compute statistics using a pre-aggregated capture file.
```

This makes it impossible to use `pytest-memray` for workflows that rely on detailed allocation statistics.

## Describe the Solution You'd like

Add a configurable option to allow users to disable aggregation and output a full (non-aggregated) memray capture file, equivalent to running `memray run` without `--aggregate`.

Possible interfaces:

- A pytest CLI flag (e.g. `--memray-no-aggregate` or `--memray-full`)
- A pytest.ini / pyproject.toml option
- An environment variable for CI use

Default behavior could remain unchanged (aggregated captures), preserving current performance characteristics.

**Drawbacks:**

- Full capture files are larger and may increase disk usage and runtime overhead.
- Users should explicitly opt in and understand the trade-off.

## Describe Alternatives You've Considered

Use old `pytest-memray` release (< 1.6.0)

## Teachability, Documentation, Adoption, Migration Strategy

- Documentation can explain the difference between aggregated and full capture modes, including which memray reporters require full allocation data.
- Examples can show how to enable full capture mode when `memray stats` or similar tools are needed.
- No migration impact: existing users see no behavior change unless they opt in.
- This option would make `pytest-memray` more flexible and suitable for advanced memory analysis workflows.

Start by tracing the pytest CLI and configuration entry points that control capture generation, then check how the current integration invokes `memray run` and aggregation. Define an opt-in path that preserves the existing default and verify that the resulting capture works with `memray stats`; update documentation and tests for both modes.

Written by the indexing model from the issue text.