feat: type extension

[abemedia]

Add support for the extension x-ogen-type. E.g. x-ogen-type: github.com/shopspring/decimal.Decimal will result in the Go type of the schema being decimal.Decimal.

A few gotchas regarding the use:

• Types specified like this will be checked to see what interfaces they implement (e.g. github.com/ogen-go/ogen/json.Marshaler, github.com/ogen-go/ogen/json.Unmarshaler, json.Marshaler, json.Unmarshaler, encoding.TextMarshaler, encoding.TextUnmarshaler, encoding.BinaryMarshaler, encoding.BinaryUnmarshaler) and will be (un)marshaled using these interfaces. If they don't implement any of them it will fall back to using the Marshal and Unmarshal functions from stdlib json package.
• If different external types have the same package name they will be aliased by adding integers e.g. decimal, decimal2, decimal3 etc.
• If multiple types have the same name, creating generics (optional, nullable) will fail due to the naming conflicts and require using x-ogen-type.
• Pointers are not currently supported. For now a pointer type (e.g. *pkg.Type) will result in a non-pointer (e.g. pkg.Type). I made an attempt at adding support for pointers but it got pretty complex pretty quickly and it wouldn't be possible to simply hijack PrimitiveType to achieve this. I did also try wrapping the type in ir.Pointer but for optional pointers this results in a pointer to a pointer (e.g. **pkg.Type) which I feel is just as bad as using a value instead of a pointer. If you think this is a requirement I can add a new type KindExternal with custom templates to handle this.
• Validation is not supported for external types.
• I added support for builtin types (e.g. string, int etc.) but on second thought, do we even want this? All builtin types can be generated using standard schema.

This PR also fixes an issue where setting default values for aliases was broken if the alias had a format (e.g. date-time, ip etc.)

Closes #1335

[ghost]

LGTM in general, but encoding is tricky, so we need a bit more tests.

Types from aliased packages will contain the name of the alias in the derived types e.g. OptDecimal1Decimal. For unique package names they will work the same as other types e.g. OptDecimal. Happy to remove this functionality though and resort to forcing users to use x-ogen-name when they have clashes if this feels too strange.

Let's remove it, then.

[abemedia]

@tdakkota I made a bunch of changes and improved the (un)marshalling of external values to avoid the reflection cost in the stdlib json package where possible. Could I get another review?
I also updated the gotchas in the PR message so I'd reread it to make sure you're happy with them.

[abemedia]

Actually hold off I just found a bug. Will fix and report back when done.

[abemedia]

@tdakkota I've resolved the bugs mentioned in the previous message and this is again ready for review.

[abemedia]

@tdakkota sorry for the delay but I have just added another commit to improve the encode/decode by removing the pointless string functions and added support for binary types.
Mind giving it another review? If you're happy I'll squash the commits, just thought this makes it easier to review the actual changes.

[abemedia]

@tdakkota I've made the final small changes and squashed the comments so it's ready for merging unless you have any other concerns.

[ghost]

Well, it was a long road.

Huge thanks for all your contributions to the ogen.

[abemedia]

Thank you!