Fix peak_time estimation in FlashCamExtractor

[clara-escanuela]

After the extractor optimization, I did not check timing performance again as I only made a few changes in the code. But one of these very small changes caused a problem in the time reconstruction. The unit tests did not identify it because they were not sensitive enough. By adding a new toymodel with a time gradient we can find timing bugs better. I checked the test with the previous implementation (with that wrong line of code) and the test fails (as expected). In fact, the same test with the standard toymodel provided (random true times) passes even when time reconstruction is wrong. I think a further test can also be added to other extractors, to avoid similar problems.

[maxnoe]

Before:

After:

[maxnoe]

@Tobychev @kosack

Could we get a second quick review here? We need to have this release to restart processing on the grid