switch to Talc memory allocator

[stlankes]

In contrast to linked_list_allocator I measured a better performance. The difference is large enough to change the crate.

Result with linked_list_allocator as default allocator:

   83430 allocation attempts,   83430 successful allocations,   83430 pre-fail allocations,   58282 deallocations
            CATEGORY | OCTILE 0       1       2       3       4       5       6       7       8 | AVERAGE
---------------------|--------------------------------------------------------------------------|---------
     All Allocations |      186    5500   14872   27390   44238   65712   97346  174924 2786786 |   84098   ticks
Pre-Fail Allocations |      186    5500   14872   27390   44238   65712   97346  174924 2786786 |   84098   ticks
       Deallocations |      180    3584    9824   18124   29224   43996   67114  119002  531484 |   53099   ticks

Result with talc as default allocator:

  571170 allocation attempts,  571170 successful allocations,  571170 pre-fail allocations,  402241 deallocations
            CATEGORY | OCTILE 0       1       2       3       4       5       6       7       8 | AVERAGE
---------------------|--------------------------------------------------------------------------|---------
     All Allocations |      250     364     568     612     668     770     876    157410049592 |   16094   ticks
Pre-Fail Allocations |      250     364     568     612     668     770     876    157410049592 |   16094   ticks
       Deallocations |      212     830    1016    1058    1204    1318    1624    1802   26080 |    1286   ticks

I used the benchmark hermit-os/hermit-rs#443, which runs 3 seconds.

Closes #630.

[mkroening]

Looks good to me! 🥳

We should update the Cargo.lock independently, though. I opened #830.