upgrade `pelias/parser` to latest version

[missinglink]

this PR upgrades the pelias/parser version to latest https://github.com/pelias/parser/releases

[Joxit]

Hi there, I just tried your PR, I think there are no ideal solution here. We are using phrase.* that contains the full address (number + street), since we need to deal with "junk in house number data" and try to match as much as data we can, stripping the alpha is for now the best solution.

Maybe we could also create some post-processing in pelias model to include alias or sanitised names/phrases mixed to a fallback query ?

I have in mind 1 bis, the data may have either 1b, 1 bis or 1bis, in the current state, 1 bis will not match 1bis. So the model post-processing may replace all 1bis to 1 bis (or add an alias). 1b is a bit tricky since it can be a unit, but as an alias of 1b it should be fine ? The final state might be {"name": {"default": ["1bis Avenue Amélie", "1 bis Avenue Amélie"]}}

Then the query we make could include a should with a minimum_should_match that will query both 1 bis Avenue Amélie and 1 Avenue Amélie as a fallback.

Oh I forgot the case when the user's input is 1bis Avenue Amélie => change to 1 bis Avenue Amélie and if it's 1b Avenue Amélie => IDK if it's safe to include 3 queries, 1b Avenue Amélie, 1 bis Avenue Amélie and 1 Avenue Amélie

It is still an idea and should not be put in this PR, this might be a big change.