Support Wasm Bulk Memory Operations

[luc-blaeser]

Use WebAssembly bulk memory operations in the runtime system and compiler.

Performance Improvement:
Although Wasm bulk memory operations are mainly beneficial for the IC execution layer, one may notice a performance improvement for the copying GC at the IC cost model.

Runtime measurements, GC benchmark, total Wasm instructions, dfx 0.14.3, with latest bump allocator optimization (only applied to non-incremental GCs):

Configuration	Before	With bulk-memory op	Difference
Copying GC	2.02e10	1.92e10	-5%
Incremental GC	1.85e10	1.85e10	0
Generational GC	1.86e10	1.86e10	0
Compacting GC	2.16e10	2.16e10	0

Fallback: ic-ref and ic-wasm do not support Wasm bulk memory operations and thus engage classical libc memcpy with ordinary memory access instructions.

Design:

• Bulk memory operations are implemented in the compiler backend. For all platforms, except for ic-ref and ic-wasm, the compiler generates Wasm bulk-memory operations.
• The runtime system uses these compiler-generated memory operations.
• If Wasm bulk memory operations are not supported (in the case of ic-ref or ic-wasm), the compiler invokes the legacy operations exposed by the RTS. For this reason, the RTS is not compiled with Wasm bulk memory flag, such that it still integrates the legacy libc implementations that do not rely on Wasm bulk memory support.

[ggreif]

See also #3009. But that needs quite some yak-shaving before it can be made working.

[luc-blaeser]

See also #3009. But that needs quite some yak-shaving before it can be made working.

Thanks for pointing this out. This PR covers changes from the former PR.

[github-actions]

Comparing from 47fb734 to ff88f8e:
In terms of gas, 4 tests regressed, 1 tests improved and the mean change is -0.2%.
In terms of size, 2 tests regressed, 3 tests improved and the mean change is -0.0%.

[ggreif]

Looks like you need some --enable-bulk-memory flags in the test/run.sh script.

[luc-blaeser]

Inactive for a while. We can reopen if needed...