This is a follow-up for !12362 where this was documented but the code
was removed in the last iteration. Since this can still be useful and
this is already supported by the API, I think re-adding the code was
the best course of action.
Signed-off-by: NRémy Coutable <remy@rymai.me>

The implementation now simply rely on the
`performance_bar_allowed_group_id` Application Setting.
Signed-off-by: NRémy Coutable <remy@rymai.me>

Signed-off-by: NRémy Coutable <remy@rymai.me>

Closes #34278

This ensures we make maximum one call per feature per request.

See
https://github.com/jnunemaker/flipper/blob/v0.10.2/docs/Optimization.md#memoizing-middlewareSigned-off-by: NRémy Coutable <remy@rymai.me>