`signTypedData` input validation

We don't have much in the way of input validation for `signTypedData`. This is dangerous because invalid inputs can result in non-standard encodings.

Some examples of this are:
* Negative `uint` values are currently encoded as positive values. The negative is ignored.
* Some out-of-bounds integer values and byte values are possible to encode right now.
* We don't validate address input, so the `address` field can be used to store values far larger than an address.

We should strictly validate input, so that users don't accidentally become reliant upon non-standard and non-portable encodings.

Start at the signTypedData entry point and trace how integer, byte, and address inputs are encoded. Use the listed negative, out-of-bounds, and oversized-address examples to define rejection behavior; done means invalid inputs are rejected and valid encodings remain standard and portable. Add focused tests for each validation case.

Written by the indexing model from the issue text.