Apply session_settings to every query

[theory]

Add a new struct, ch_query, and replace all instances of passing a char * with passing ch_query. This allows it to included additional fields relevant to the query, starting here with sql and settings, the latter currently being a PostgreSQL List of the settings parsed from pg_clickhouse.session_settings. Provide new_query() to keep its construction in one place; in the future it might take parameters to merge additional settings fetched from a server or user mapping to merge into the GUC's settings.

Update ch_binary_simple_query() to iterate over the list of settings in the ch_query and apply them to the query as a QuerySettings map. Add a comment to do the same for inserts, held until clickhouse-cpp#453 is merged and released.

Update ch_http_simple_query() to use curl_url_set() to build each request URL, starting with the base URL and query ID, then adding another query parameter for each item in ch_query.settings. This also eliminates the need for ch_http_connection_t.base_url_len.

Add tests to ensure consistent application of a variety of settings via both the binary and http engines.

While at it, fix a bug where http query results treated an empty string as NULL and add a test for it. Also replace a buffer allocation with simple use of psprintf(), which is automatically cleaned up in the current PostgreSQL memory context.

Other bits:

• Document pg_clickhouse.session_settings
• Rename connect.h to engine.h, which better describes its purpose providing objects and functions used by both the http and binary engines
• Remove the unused cursor_free_method field from the libclickhouse_methods struct

[iskakaushik]

Nice refactor. Few things:

1. http.c - the psprintf() calls in the foreach loop for URL params should be pfree()'d
   after each curl_url_set(). Same for the query_id one. Curl copies the string so we can free
   immediately.
2. kill_query() lost its pfree() - the old code freed the query string, new code doesn't.
   Might need to inline the macro instead of using new_query() to get the sql pointer back for
   freeing.
3. chfdw_construct_create_tables() reassigns query.sql in a loop without freeing previous -
   minor but adds up.

Nits:

• engine.h comment says "ch_connection_details" but should say "ch_query"
• The (List *) cast in binary.cpp:180 is fine but a one-liner comment explaining it's
  because foreach needs non-const would help future readers

Rest looks good, especially the empty string fix and using curl's URL API.

[theory]

Nice refactor. Few things:

1. http.c - the psprintf() calls in the foreach loop for URL params should be pfree()'d
   after each curl_url_set(). Same for the query_id one. Curl copies the string so we can free
   immediately.

Makes sense.

2. kill_query() lost its pfree() - the old code freed the query string, new code doesn't.
   Might need to inline the macro instead of using new_query() to get the sql pointer back for
   freeing.

It shouldn't need to pfree(); the memory will be freed when the memory context is freed. That's the benefit of using palloc() and friends.

3. chfdw_construct_create_tables() reassigns query.sql in a loop without freeing previous -
   minor but adds up.

Check.

Nits:

• engine.h comment says "ch_connection_details" but should say "ch_query"
• The (List *) cast in binary.cpp:180 is fine but a one-liner comment explaining it's
  because foreach needs non-const would help future readers

Rest looks good, especially the empty string fix and using curl's URL API.

Thanks!

[theory]

Fixed @iskakaushik's issues in b015f06 (diff).