Refactor XDR serialization/deserialization logic

[orbitlens]

• Refactor XDR serialization/deserialization logic.
• Replace long dependency with native BigInt arithmetics.
• Replace lodash dependency with built-in Array and Object methods, iterators.
• Add buffer dependency for WebPack browser polyfill.
• Update devDependencies to more recent versions, modernize bundler pipeline.
• Automatically grow underlying buffer on writes (#84 fixed).
• Always check that the entire read buffer is consumed (#32 fixed).
• Check actual byte size of the string on write (#33 fixed).
• Fix babel-polyfill build warnings (#34 fixed).
• Bump major version.

---

Minified browser bundle size decreased from 335 KB (prev version) to 43 KB (this version). Memory consumption during serialization decreased due to the smaller writer buffer allocation.

We already actively use it as a drop-in replacement for original js-xdr package for ingestion. So far, no compatibility issues have been found.

Also I created a primitive performance benchmark for Node.js to test serialization/deserialization speed for refactored version vs original. It shows significant performance boost, up to 5X depending on the serialized type.

Benchmark link. Benchmark results on my computer:

Performance benchmark
10 samples 10000 iterations each
------------------------------------------------
LedgerHeader.fromXDR - new: 0.0435s (min 0.038s - max 0.068s)
LedgerHeader.fromXDR - old: 0.272s (min 0.264s - max 0.302s)
difference: 6.3X

LedgerHeader.toXDR - new: 0.0698s (min 0.06s - max 0.101s)
LedgerHeader.toXDR - old: 0.348s (min 0.263s - max 0.558s)
difference: 5.0X

TransactionEnvelope.fromXDR - new: 0.423s (min 0.412s - max 0.477s)
TransactionEnvelope.fromXDR - old: 1.77s (min 1.701s - max 1.918s)
difference: 4.2X

TransactionEnvelope.toXDR - new: 0.436s (min 0.342s - max 0.715s)
TransactionEnvelope.toXDR - old: 0.646s (min 0.613s - max 0.695s)
difference: 1.5X

TransactionMeta.fromXDR - new: 3.20s (min 2.694s - max 4.016s)
TransactionMeta.fromXDR - old: 9.76s (min 9.543s - max 10.194s)
difference: 3.1X

TransactionMeta.toXDR - new: 2.35s (min 2.149s - max 2.549s)
TransactionMeta.toXDR - old: 3.19s (min 2.934s - max 3.46s)
difference: 1.4X

TransactionResult.fromXDR - new: 0.0370s (min 0.034s - max 0.056s)
TransactionResult.fromXDR - old: 0.0627s (min 0.059s - max 0.074s)
difference: 1.7X

TransactionResult.toXDR - new: 0.0626s (min 0.057s - max 0.07s)
TransactionResult.toXDR - old: 0.218s (min 0.184s - max 0.272s)
difference: 3.5X

[orbitlens]

Wow, Herculean stuff 👏 It'd be nice if we had .x -> .js tests to see if there were any output changes.

I tried generating .x -> .js with examples/test.x (via yarn run test-generate), but of course it makes sense to test with .x files from stellar-core. Could you please check the generated output for the difference? Because I'm a bit lost between various versions of stellar-core and curr/next XDR.

[tomerweller]

Great work. thanks @orbitlens!