Use ahash by default

[ritchie46]

ahash is a dos resistant hashing algorithm that is much faster than std's default. Given that there are users with millions of columns, not paying unneeded cryptographic hashing costs seems worth it to me.

Aside from the crate being pulled in, this should not have much effect on build times as the hasmaps/sets get compiled on instantiation.

[codecov]

Codecov Report

Merging #1148 (211e62a) into main (e4947cc) will decrease coverage by 0.00%.
The diff coverage is 90.47%.

@@            Coverage Diff             @@
##             main    #1148      +/-   ##
==========================================
- Coverage   83.62%   83.62%   -0.01%     
==========================================
  Files         366      366              
  Lines       35906    35911       +5     
==========================================
+ Hits        30028    30031       +3     
- Misses       5878     5880       +2     

Impacted Files	Coverage Δ
src/array/union/mod.rs	77.41% <ø> (ø)
src/io/avro/mod.rs	86.66% <ø> (ø)
src/io/ipc/read/reader.rs	96.65% <0.00%> (+0.66%)	⬆️
src/lib.rs	100.00% <ø> (ø)
src/compute/like.rs	41.29% <50.00%> (ø)
src/io/ipc/read/common.rs	94.92% <90.00%> (-0.30%)	⬇️
src/compute/merge_sort/mod.rs	95.60% <100.00%> (ø)
src/compute/regex_match.rs	82.35% <100.00%> (ø)
src/io/avro/read/header.rs	100.00% <100.00%> (ø)
src/io/avro/read/util.rs	96.29% <100.00%> (ø)
... and 11 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1f2116e...211e62a. Read the comment docs.

[jorgecarleitao]

AFAIK std is already using ahash, so there isn't much to gain performance-wise. I think the benefit of ahash is a public API related to entry?

[ritchie46]

AFAIK std is already using ahash, so there isn't much to gain performance-wise. I think the benefit of ahash is a public API related to entry?

The default hasher is SipHash 1-3. See std. I know that hashbrown is using ahash, but not std.

Quoting:
"The default hashing algorithm is currently SipHash 1-3, though this is subject to change at any point in the future".

Here are some benchmarks of siphash vs ahash, showing that it's ~10x faster across all dtypes. /~https://github.com/tkaitchuck/aHash/blob/master/FAQ.md

[ritchie46]

Rebased. 👍