Improve weightedmatching

[Matistjati]

Update our code for weighted matching.
The new one:

• Makes it easy if costs are non-int (long long or double). int is way too small
• Does not require you to specify an infinity for your type
• Runs faster (on the problem I've tested on)
• Fewer LOC (24 vs 32). Slightly more non-whitespace chars: went from 556 to 601

This code was produced by taking Jeremy Lim's implementation of Jonker-Volgenant, and then getting LLMs to golf it for a couple of hours. There might have been some performance loss from this.

Benchmarks. The ones other assignment problem use an older version that should compile the same. Take the exact numbers with a grain of salt, the variance seems to quite large:

• Yosupo Assignment Problem. 131 ms -> 81 ms
• Kattis Engaging with Loyal Customers. 1.53s -> 1.02s
• Kattis Cordon Bleu. 1.67s -> 0.99s
• Kattis Aqueduct Construction.0.21s -> 0.21s

[simonlindholm]

I think we should keep allowing n = 0, it feels like it comes up quite naturally. rep(i,f,t) rather than rep(i, f, t) to keep with existing style. Not super happy about the unnatural line breaks or overuse of comma expressions. Shortening potential might allow some of the line breaks to be fixed maybe.