Distinguish (ACT + historical_activity) × output from `out` in reporting

During the ScenarioMIP/SSP project, @OFR-IIASA and @macflo8 reported the following issue using:
1. `Scenario.clone(…, shift_first_model_year=…)`.
2. `message_ix.report` output.
3. `message_data.tools.post_processing` (~`message_ix_models.report.legacy`) with the `merge_ts`/`merge_hist` options.

In essence what happens:
- After (1), for the set of “newly historical” periods—that is, those that *were* originally within the model horizon but *become historical periods* due to the shift—the Scenario is left with (a) zeros in `ACT`, (b) non-zero values in `historical_activity`, and (c) non-zero values in `output`. For example, if y₀=2020 initially but `shift_first_model_year=2030` is given, then the affected periods might be 2020 and 2025.
- In (2), the calculation of `out:* = ACT:* × output:*` thus yields zeros for those periods.
- In (3), those zeros overwrite existing/historical values for those periods.

Users will almost never want to be using this quantity with usable values for the new y₀ and later combined with zeros for the newly-historical periods.

Instead, the **expected behaviour** is that the user can make a deliberate and informed choice to use either of the following:
- `out:* = ACT:* × output:*`, but *only for the periods y₀ and later*. This would omit the zeros produced for the newly-historical periods.
- `out:*:hist = (ACT:* + historical_activity:*) × output:*`. This would add the `historical_activity` values for those newly-historical periods to the `ACT` zeros, leading to a non-zero values in the result. This mirrors the way the variable `ACT` and parameter `historical_activity` are used in the GAMS code.

The tag `:hist` is tentative, perhaps there is a more intuitive and informative way to make the distinction.

Trace the flow from Scenario.clone through message_ix.report and message_data.tools.post_processing, including message_ix_models.report.legacy and the merge_ts/merge_hist options. Compare the two proposed output choices for newly historical periods, then confirm the behavior and naming make the distinction explicit without overwriting historical values unexpectedly.

Written by the indexing model from the issue text.