add InfluxDB 3.x support

[crandles]

InfluxDB 3.x (GA April 2025, now at 3.9.1) is a ground-up rewrite -- Flux is gone, SQL is the primary query language, and there are new /api/v3/ endpoints for queries and writes. Trickster's existing InfluxQL support works against v3 via compat endpoints, but there's no caching for SQL queries or the native v3 API.

Main gaps: SQL query caching (/api/v3/query_sql), v3 InfluxQL endpoint (/api/v3/query_influxql), and new response formats (jsonl, csv). Can model the SQL time-range parsing after the clickhouse backend.

Flight SQL (gRPC)

v3 also exposes Apache Arrow Flight SQL on gRPC. Grafana's InfluxDB datasource in SQL mode uses Flight SQL, not HTTP -- so without this, SQL panels in Grafana can't go through Trickster. Other high-perf v3 clients (Python/Rust/Java SDKs) also default to Flight SQL.

To support this we need:

• gRPC listener (extend the non-HTTP protocol listener pattern from feat: [clickhouse] native binary protocol + parser + grafana v4 plugin #981, or a new gRPC-aware listener variant -- grpc.Server.Serve(lis) owns the whole listener, unlike raw-TCP per-conn handling)
• Flight SQL server implementation -- handle CommandStatementQuery + DoGet
• Apache Arrow Go (github.com/apache/arrow-go/v18) for Arrow RecordBatch serialization
• Reuse existing SQL parser + delta-proxy cache by querying upstream over HTTP SQL (/api/v3/query_sql?format=json), converting to Arrow on the way out. Upstream-via-Flight-SQL is a later optimization.
• Cache interop: dataset.DataSet round-trips to Arrow record batches

Tracking as part of this issue since the goal is the same (cache v3 SQL). Scope may split into separate PRs.