Fix integer overflow on 32-bit architectures

[chrislusf]

Problem

On 32-bit architectures (such as OpenBSD ARM, Linux ARM, etc.), building the parser fails with integer overflow errors:

pkg/util/tsearch/eval.go:180:11: cannot use math.MaxInt64 (untyped int constant 9223372036854775807) as int value in assignment (overflows)
pkg/sql/sem/tree/format.go:185:2: FmtFlags(lexbase.EncFirstFreeFlagBit) << iota (constant 2147483648 of int type FmtFlags) overflows int

The issue occurs because:

1. int type is 32-bit on 32-bit architectures
2. Variables declared as int cannot hold 64-bit values like math.MaxInt64
3. Bit shift operations that produce values > 32 bits overflow

Solution

This PR changes the affected types from int to int64:

1. pkg/util/tsearch/eval.go: Changed lPos and rPos variables from int to int64 to accommodate math.MaxInt64
2. pkg/sql/lexbase/encode.go: Changed EncodeFlags type from int to int64 to support large bit shift operations
3. pkg/sql/sem/tree/format.go: Changed FmtFlags type from int to int64 for the same reason

Testing

Successfully built and tested on:

• GOOS=openbsd GOARCH=arm (32-bit)
• Standard 64-bit platforms (Linux amd64, Darwin arm64)

The changes are backward compatible since int64 is a superset of int on all platforms.

Impact

• Minimal performance impact (int64 operations are native on all modern CPUs)
• Enables builds on 32-bit architectures
• No API changes or breaking changes