Reduce dependency bloat

[Kixiron]

• Started trying to remove num from the dependency tree
• Removed some of the "default dependencies" from generated crates, this should allow better compilation scheduling since crates don't have to wait on unnecessary dependencies
• Discovered that some files are in CLRF form, this will be fixed in a later PR
• Removed dependencies on fnv and twox-hash and replaced with a single dependence on xxhash-rust, it has zero dependencies and is very light
• Removed old dependence on time for all generated crates, this reduced the dependency tree a good bit
• Minor refactors to some massive match statements
• Added some documentation to the DDVal/DDValue innards

[mihaibudiu]

Some of these files show every line as different. Is this due to CR/LF?
This makes it hard to review.

[Kixiron]

Yes, those are files that were changed from CLRF to LF, I'll make a PR that bites the bullet and standardizes things after this.
In the diff view select the Hide Whitespace option to ignore the line ending changes

[ryzhyk]

There's a bunch of test failures. Some of them are probably caused by full disks on CI machines. Those should now be fixed. Others probably just require changing reference test outputs due to the use of a different hashing function.

I think most of the maps that use FNV today can be regular default HashMaps. I mostly used FNV to ensure deterministic iteration order. This is necessary e.g., when building dataflow graphs to make sure that each DD worker ends up with the same operator id assignment. In most other cases there's no reason for non-standard hasher.

[ryzhyk]

why don't we trust the compiler with all this inlining?

[Kixiron]

We do, however this makes the functions viable for cross-crate inlining