Remove OpenSnapshot and CommittedSnapshot markers.

[nnethercote]

They're not strictly necessary, and they result in the Vec being
allocated even for the trivial (and common) case where a
start_snapshot is immediately followed by a commit or rollback_to.

This commit removes them, and instead uses a counter to track how many
snapshots are open. One consequence of this is that violations of stack
discipline in API usage may be detected later -- e.g. if we rollback to
an earlier snapshot and then to a later snapshot, the code will panic on
the second rollback instead of the first. I think this is an acceptable
trade-off for the improved performance.

This commit also adds some basic unit tests.

[nnethercote]

I removed these markers in ena and also in two places in rustc itself (librustc_data_structures/snapshot_map/mod.rs and librustc/infer/region_constraints/mod.rs). With all the changes in place, the instruction count results were remarkably good:

futures-check
	avg: -3.9%	min: -6.2%	max: -1.2%
packed_simd-check
	avg: -3.3%	min: -5.4%	max: -0.4%
serde-check
	avg: -2.7%	min: -3.7%	max: -0.8%
piston-image-check
	avg: -2.1%	min: -3.5%	max: -0.4%
crates.io-check
	avg: -2.1%	min: -3.3%	max: -0.5%
deep-vector-check
	avg: -2.1%	min: -3.2%	max: -0.1%
tokio-webpush-simple-check
	avg: -2.2%	min: -3.1%	max: -0.5%
tuple-stress-check
	avg: -1.8%	min: -3.0%	max: 0.3%
unify-linearly-check
	avg: -2.4%	min: -2.9%	max: -1.9%
ripgrep-check
	avg: -1.7%	min: -2.8%	max: -0.3%
regex-check
	avg: -1.2%	min: -2.8%	max: -0.4%
regression-31157-check
	avg: -2.0%	min: -2.7%	max: -1.0%
coercions-check
	avg: -1.4%?	min: -2.7%?	max: 0.4%?
webrender-check
	avg: -1.5%	min: -2.7%	max: -0.2%
sentry-cli-check
	avg: -1.5%	min: -2.4%	max: -0.2%
ucd-check
	avg: -1.5%	min: -2.4%	max: -0.5%
encoding-check
	avg: -1.5%	min: -2.3%	max: -0.5%
helloworld-check
	avg: -2.1%	min: -2.3%	max: -2.0%
syn-check
	avg: -1.5%	min: -2.3%	max: -0.4%
ctfe-stress-check
	avg: 0.1%?	min: -1.5%?	max: 2.2%?
cargo-check
	avg: -1.3%	min: -2.2%	max: -0.1%
style-servo-check
	avg: -1.4%	min: -2.0%	max: -0.1%
clap-rs-check
	avg: -1.4%	min: -2.0%	max: -0.8%
issue-46449-check
	avg: -1.7%	min: -1.9%	max: -1.4%
deeply-nested-check
	avg: -1.8%	min: -1.9%	max: -1.6%
html5ever-check
	avg: -1.1%	min: -1.9%	max: -0.0%
rust-unic-check
	avg: -1.0%	min: -1.2%	max: -0.9%
unused-warnings-check
	avg: -0.0%	min: -0.2%	max: 0.4%
inflate-check
	avg: -0.3%	min: -0.4%	max: -0.3%
keccak-check
	avg: -0.3%	min: -0.4%	max: -0.1%

[nnethercote]

I guess a new version of ena would need to be published for rustc to be able to pick up these changes. @nikomatsakis: Should I bump the version number to 0.11.0 in a second patch?

[nnethercote]

@nikomatsakis: any thoughts? This is part of a good-sized win.

[nnethercote]

I removed these markers in ena and also in two places in rustc itself (librustc_data_structures/snapshot_map/mod.rs and librustc/infer/region_constraints/mod.rs).

The rustc changes are available in rust-lang/rust#55906.

[nikomatsakis]

@nnethercote sorry, didn't see your pings till just now -- I'm 👍 on ena changes.