Fix charset-aware MySQL expressions and result encoding

[ketor]

Follow-up to #1853 (merged before the review fixes could be added).

• Correct CHAR numeric rounding/decimal-text truncation and reject invalid byte values.
• Return charset-aware VARCHAR from CONVERT(... USING ...) and CHAR(... USING ...), preserving strict Windows-1252 (latin1) and ASCII conversion rather than disguising text as binary.
• Carry operand charset through CONCAT and use it for byte-sensitive LENGTH/HEX; expose expression charset in JDBC metadata. Promote mixed text without losing characters: Unicode wins over non-Unicode at equal coercibility, representable literals retain the expression charset, and ASCII can promote to latin1.
• Encode text and prepared result rows in the negotiated character_set_results, advertise the matching wire collation, reject unrepresentable characters, and correct NOT_NULL and CONVERT_TZ nullable metadata.

Verification on the 172.30.14.10 QA executor after deploying the final candidate:

• ./gradlew :dingo-calcite:test :dingo-driver:dingo-mysql-service:test :dingo-test:test :dingo-test:compileIntTestJava --console=plain — BUILD SUCCESSFUL. The new mixed-charset regression failed before the fix and passes afterward; the existing latin1+literal regression still passes.
• Live MySQL CLI: latin1 € + utf8mb4 🙂 yields LENGTH=7, HEX=E282ACF09F9982; reversed yields F09F9982E282AC; three-operand concat yields LENGTH=9, HEX=E282ACF09F9982C2A3. Both ASCII/latin1 orders preserve the euro (8041 / 4180). Direct result rows decode as €🙂 and €A under utf8mb4 result collation 45.
• Live Connector/J 8.0.33 server-prepared parameterized concat returns €🙂, HEX=E282ACF09F9982, Types.VARCHAR.
• Earlier live checks: MySQL CLI UTF-8 results €, é; Connector/J direct and prepared VARCHAR text, nested LENGTH(CONCAT(...)) = 2, latin1 euro HEX = 80, numeric CHAR, invalid ASCII rejection, nullable invalid-zone CONVERT_TZ, and both SET NAMES latin1 / SET character_set_results=utf8mb4 result modes.

Additional fix bd854ca (mixed-charset CASE): SQL validation and Rex conversion now preserve Unicode regardless of branch order, including nullable ELSE branches; direct CASE projection metadata reports nullable correctly. QA verification after the change: ./gradlew :dingo-calcite:test --tests io.dingodb.calcite.TestDingoSqlValidator --tests io.dingodb.calcite.TestRex :dingo-test:test --tests io.dingodb.test.QuerySimpleExpressionTest --console=plain — BUILD SUCCESSFUL. Restarted QA executor live MySQL query returned F09F9982, LENGTH=4, reverse-case E282AC, and nullable-case F09F9982.