CBOR: bignum payloads bypass bounded_bytes chunking above 64 bytes

[solidsnakedev]

Summary

Plutus Data integers whose magnitude exceeds 64 bytes (|value| >= 2^512) are encoded as a single
definite-length byte string under CBOR tag 2 or 3. The Conway CDDL, quoted in this repo at
Data.ts L165-167, defines a bignum payload as bounded_bytes = bytes .size (0..64), so a magnitude
of 65 bytes or more must be an indefinite-length byte string split into chunks of at most 64 bytes.
The SDK already applies that rule to Data byte-string leaves, so the same file chunks a 65-byte
bytearray and does not chunk a 65-byte bignum. The result is Plutus data that violates the grammar,
and a datum hash that differs from the one computed over the chunked form.

Affected

packages/evolution/src/CBOR.ts

• encodeUintSync (L985-991): passes bigintToBytes(value) straight to encodeTagSync(2, ...) (L989-990)
• encodeNintSync (L1028-1035): same for tag 3 (L1033-1034)
• contrast: the BoundedBytes node (L1231-1239) routes through encodeBoundedBytesSync (L1109),
  which chunks at 64 bytes and is applied unconditionally

packages/evolution/src/Data.ts

• CDDL for big_uint / big_nint / bounded_bytes documented at L165-167

Fix

Wrap the bignum magnitude in the existing BoundedBytes node so the chunking rule applies, mirroring
the fix PR #160 made for byte-string leaves:

return encodeTagSync(2, BoundedBytes.make(bigintToBytes(value)), options, fmt)

and the tag 3 analogue in encodeNintSync. Alternatively have encodeTagSync treat tag 2 and 3
payloads as bounded bytes. Note #395 edits the same two lines of encodeNintSync for the negative
bignum boundary, so the two changes will conflict.

Regression test

• given: Data.int(2n ** 512n)
• before fix: encodes to c25841... (single 65-byte definite string); toDatumHash is 9b2838f7...
• after fix: encodes to c25f5840...4100ff; toDatumHash is
  939bb1f51ea88f86a070fefb5c1185d10b4d10a13e3b7db24aac02220c18adab
• negative analogue: Data.int(-(2n ** 520n)) chunks under tag 3 rather than emitting c35841...
• unchanged: a 9-byte bignum such as 2n ** 64n stays definite, so only the over-64-byte case moves

Must FAIL on main today and PASS after the fix.

Reference

Same defect class as #158, which covered Data byte-string leaves and was fixed in PR #160. The
integer path was never routed through BoundedBytes.