[R][C++] Why is read_csv_arrow so much slower when using S3 path notation?

Consider these two mechanisms for reading from a public bucket.  I was struck to see that using S3 path notation was consistently over 20 times slower than using the https address directly.  I could imagine a small overhead for using S3, but compared to other operations this seems something weird is going on here:

```java

library(arrow)
targe <- s3_bucket("neon4cast-targets", endpoint_override="data.ecoforecast.org", anonymous=TRUE)

bench::bench_time({ # 58.6 seconds
  ex1 <- read_csv_arrow(targe$path("terrestrial_30min/terrestrial_30min-targets.csv.gz"))
})

bench::bench_time({ # 2.7 sec
  ex2 <- read_csv_arrow("https://data.ecoforecast.org/neon4cast-targets/terrestrial_30min/terrestrial_30min-targets.csv.gz")
})

```

**Reporter**: [Carl Boettiger](https://issues.apache.org/jira/browse/ARROW-17597) / @cboettig

_{**Note**: *This issue was originally created as [ARROW-17597](https://issues.apache.org/jira/browse/ARROW-17597). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Reproduce the two bench_time cases using read_csv_arrow with the s3_bucket path and the direct HTTPS URL. Start at the read_csv_arrow and s3_bucket entry points, compare their request and decompression behavior, and use the benchmark timings to confirm that the S3-path performance gap is resolved.

Written by the indexing model from the issue text.