Refactor address_transform.rs to use less memory

[yurydelendik]

The crates/debug/src/transform/address_transform.rs is unoptimized in terms of data structures. This PR refactors this file to remove creation of intermediate in-heap structures, thus improves overall performance of the DWARF transformation.

[fitzgen]

Looks good to me! A few ideas for follow up PRs in the comments below.

[fitzgen]

I wonder if we could make this a u32 in practice? Do we ever have to process Wasm files larger than 4GiB?

This would let us shave off half the overhead on 64-bit platforms.

[fitzgen]

Rather than having our index be a BTreeMap<WasmAddress, Box<[RangeIndex]>, which implies a bunch of allocated boxed slices that are all relatively small and potentially contribute to internal and/or external memory fragmentation, we should be able to save some memory by doing:

index_ranges: Vec<RangeIndex>,
// A map from the wasm address to the range within `index_ranges` for the associated wasm address.
index: BTreeMap<WasmAddress, std::ops::Range<usize>>,

Also if we can get away with it, we could make it a Range<u32> for extra savings.

To do this, rather than inserting sorted_ranges into the index here, we would do something like:

let start = all_ranges.len();
let end = start + sorted_ranges.len();
index_ranges.extend(sorted_ranges);
index.insert(position, start..end);