Replace contour HashMap keys with integer tuples instead of Strings

[antbern]

Some small changes to improve execution time and memory usage for the xyz2contours function even further!

• Changed the keys in the obj and curves HashMap to be tuples of (i64,i64,u8) instead of strings that needed to be formatted and split on every iteration. Since this is quite a hot code-path it improves both memory usage (needed allocations on the heap) and runtime (less operations to perform) 🚀
• Changed to use the more efficient read_lines_no_alloc function when converting the result of the curve generation.
• Improved file handling to not open and close the file on every iteration since lines were only appended anyways.
• Instead of reading in all the points and storing them for later computing the average of each cell, we now only store the sum and the count. This leads to a substantial reduction in memory usage during this part of the processing. On my tests down ca 100MB 👍

[rphlo]

This saved a bit more than 10% (1 minute) on my usual batch execution time (now at 8m25s).

Anyway, Great job @antbern !

Do you have already something in mind for next weekend?

[antbern]

This saved a bit more than 10% (1 minute) on my usual batch execution time (now at 8m25s).

Wow that is amazing! On my single-file job it cut something like 4-8 seconds. Down to 1:25 mins.

Do you have already something in mind for next weekend?

Yes I do, I started replacing all the string concatenation and splitting in knolldetector with Vec and structs which seems to have improved memory usage and speed for that quite a bit (see #110 ). Also there is a place where we read the "pins.txt" file in a loop that is quadratic in the number of rows (for computing the distances between all of them) causing something like 700*700 reads of the file 😄 Better to read the entire file to memory and then to the logic IMO 🚀