Always compress responses when the client compression is on

[zhkvia]

Summary

• It resolves An obscure error in the HTTP protocol when using compression #1150 again, since the fix was broken at some time (body after return is empty since it hasn't been read yet)
• The previous fix covered only the 403 code, but this one covers all non-200
• As a bonus, with this fix the error doesn't state that it failed to read the response (which is technically not true since a response is successfully read)
  Now it looks like:

clickhouse [execute]:: 403 code: Code: 516. DB::Exception: default: Authentication failed: password is incorrect, or there is no user with such name.
<and the rest>

[CLAassistant]

All committers have signed the CLA.

[zhkvia]

Actually after applying my fix and running it for some time in production I noticed that errors returned from a failed DESCRIBE don't get decompressed:

clickhouse [execute]:: 404 code: ~k���"x�
�h��`�������Code: 60. DB::Exception: Table logs.���_shard does not exist. Maybe you meant0�0_trace?. (UNKNOWN_TABLE) (version 23.8.12.13 (official build))

So I looked at the source code more closely and noticed the difference between DESCRIBE and INSERT queries:

1. responses from DESCRIBE queries are always compressed:
   1. https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http_batch.go#L54
   2. https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http_query.go#L39
2. while responses from INSERT queries are not:
   1. https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http_batch.go#L52
   2. https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http_batch.go#L108
   3. https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http_batch.go#L201

But responses are always decompressed no matter what a query it was: https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http.go#L432

So in order to fix this, I explicitly set compress option for INSERT queries as well

---

More than that, I noticed that in older versions of the library EOF was treated explicitly because it was wrapped with other errors several times in a row (e.g. read next block: header: EOF):
https://github.com/ClickHouse/clickhouse-go/blob/v2.14.0/conn_http.go#L453

But now this error is not checked and ReadAll can't handle it by itself since it doesn't look inside wrapped errors:
https://github.com/ClickHouse/clickhouse-go/blob/main/conn_http.go#L436

So I fixed this too

[zhkvia]

run-tests / single-node (1.21, 24.1) (pull_request) Failing after 1m

Well, I couldn't reproduce the same results locally:

 λ  CLICKHOUSE_VERSION=24.1 go test .
ok      github.com/ClickHouse/clickhouse-go/v2  0.392s
 λ  cd tests/issues
 λ  CLICKHOUSE_VERSION=24.1 go test .
ok      github.com/ClickHouse/clickhouse-go/v2/tests/issues     0.565s
 λ  go version
go version go1.21.9 darwin/amd64

Also

 λ  CLICKHOUSE_VERSION=24.1 make test

runs fine