Host user creation - Only update groups if needed

[atburke]

This change updates host user creation to only update a host user's groups when they have changed (in the role). This allows nodes to skip acquiring a user lock most of the time, which fixes performance issues when a Teleport user execs repeatedly.

Benchmarks from running tsh ssh 5 times concurrently are below.

Before:

Command	Mean [s]	Min [s]	Max [s]	Relative
seq 5 | parallel tsh ssh -l keeptemp node echo	9.338 ± 4.842	5.351	15.777	1.00

After:

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
seq 5 | parallel tsh ssh -l keeptemp node echo	367.2 ± 26.9	342.7	505.5	1.00

Changelog: Improved tsh ssh performance for concurrent execs

[strideynet]

Looks good - thanks for including some performance measurements. Definitely looks like it makes a big difference!

[atburke]

Friendly ping @fspmarshall

[atburke]

Friendly ping @fspmarshall @rosstimothy

[rosstimothy]

I'm working on a benchmark test to cover the case that identified this regression. Below is the comparison from running it on master and running it on this branch.

benchstat old.txt new.txt 
goos: linux
goarch: arm64
pkg: github.com/gravitational/teleport/lib/srv/regular
                                 │     old.txt     │               new.txt                │
                                 │     sec/op      │    sec/op     vs base                │
ExecCommand/no_user_creation-8       340.5m ±  17%   338.5m ± 10%        ~ (p=0.579 n=10)
ExecCommand/with_user_creation-8   50110.2m ± 110%   289.5m ± 20%  -99.42% (p=0.000 n=10)
geomean                               4.131          313.0m        -92.42%

[public-teleport-github-review-bot]

@atburke See the table below for backport results.

Branch	Result
branch/v14	Failed
branch/v15	Create PR
branch/v16	Create PR

[rosstimothy]

Regression originally introduced by #41919