zap logger adapter should not do more work than neccessary

zap logger adapter does a lot of work massaging fields without even checking if a particular logging level is enabled. A more efficient implementation will look something like the following:

```go
func (pl *Logger) Log(ctx context.Context, level pgx.LogLevel, msg string, data map[string]interface{}) {
zapLevel := zapcore.DebugLevel
keepSourceLevel := false

switch level {
case pgx.LogLevelInfo:
zapLevel = zapcore.InfoLevel
case pgx.LogLevelWarn:
zapLevel = zapcore.WarnLevel
case pgx.LogLevelError:
zapLevel = zapcore.ErrorLevel
default:
keepSourceLevel = true
}

ce := pl.logger.Check(zapLevel, msg)
if ce == nil {
return
}

fields := make([]zapcore.Field, 0, len(data) + 1)
for k, v := range data {
fields = append(fields, zap.Any(k, v))
}

if keepSourceLevel {
fields = append(fields, zap.Stringer("PGX_LOG_LEVEL", level))
}

ce.Write(fields...)
}
```

Locate the zap logger adapter and its Log method, then read the surrounding logging and test code to understand the current field handling. Done means disabled log levels avoid field processing while enabled levels preserve the existing message, fields, and source-level behavior; add or update tests for both paths.

Written by the indexing model from the issue text.