CLIENT-5011 server side idx selection

[gagan405]

Related: aerospike/aerospike-client-python-async#85

This enables server side index selection. The base branch is CLIENT-4878 as index selection will go along with AEL parsing.

[paul-brackin]

This will likely require another pass as it was just a lot.

In addition to all the line items.. a couple more thoughts:

• In PAC #85: Version.supports_query_selection() hard-codes true. I would give it the same dark-launch treatment as feature_gates.py here: return false (or the real core check) until the server merges the optimizer; otherwise the moment PSDK's gate flips, PAC reports support against servers that can't serve field 44.

• Naming: PSDK's supports_server_compiled_ael vs JSDK's supportsAel(). This is the one genuine naming divergence. PSDK's name is the better one as long as we still have client-side AEL. But as soon as client-side AEL goes away, JSDK's name becomes the better choice. My understanding is client-side AEL is going away & if that is true, we should align the PSDK's flag the the JSDK's; Vice-versa, if that is not true.

[gagan405]

In PAC #85: Version.supports_query_selection() hard-codes true. I would give it the same dark-launch treatment as feature_gates.py here: return false (or the real core check) until the server merges the optimizer; otherwise the moment PSDK's gate flips, PAC reports support against servers that can't serve field 44.

I will mark that as False. It was set to true as QA is testing against a server release that supports Index Optimization.
Also, server side has already merged. So, we can actually cleanup quite a lot.

Naming: PSDK's supports_server_compiled_ael vs JSDK's supportsAel(). This is the one genuine naming divergence. PSDK's name is the better one as long as we still have client-side AEL. But as soon as client-side AEL goes away, JSDK's name becomes the better choice. My understanding is client-side AEL is going away & if that is true, we should align the PSDK's flag the the JSDK's; Vice-versa, if that is not true.

Client side AEL will go away. I will align the names, when we clean up the client side parsing.

[paul-brackin]

OK other than the line items, some thoughts:

• PSDK gate logic is scattered where JSDK centralizes. Five sites each do their own flag lookup + materialize (where(), default_where(), ops helper, QueryBinBuilder.select_from, background builders). JSDK routes everything through AelMaterializer. ael/server_filter.py is the seed of exactly that; grow it into the single materializer (takes the flag source, one code path), and the eventual client-AEL deletion becomes a one-module change instead of a five-site hunt.

• Plan logging parity. JSDK debug-logs the resulting plan (CLIENT-5066: selection, index name); PSDK's debug line logs only "explain→execute" intent. Log plan.selection/index_name after explain; it's the field-diagnosis tool for exactly the probe-3 class of question ("did my query use the index?").