refactor: remove immer include from public runtime headers

[butterunderflow]

No description provided.

[butterunderflow]

The speedup of compilation time.

Case	before removing immer (s)	after removing immer (s)	speedup
ack	0.421	0.402	1.047×
branch-strip-buggy	0.419	0.403	1.038×
staged.brtable_concolic	0.401	0.378	1.062×
staged.simple_global	0.377	0.354	1.065×
mem-sym	0.772	0.743	1.039×
mem-sym-extract	0.402	0.382	1.054×
fp_ops	0.639	0.611	1.045×
btree.2o1u-unlabeled	2.891	2.848	1.015×
staged.return_poly	0.386	0.364	1.061×
pow	0.405	0.381	1.064×
start	0.369	0.346	1.065×
fact	0.407	0.380	1.071×
even_odd	0.417	0.392	1.063×
global-sym	0.389	0.372	1.047×
load	0.388	0.360	1.078×
select	0.401	0.381	1.054×
load-overflow1	0.384	0.364	1.057×
load-overflow2	0.385	0.359	1.072×
load-offset	0.386	0.368	1.050×
fib	0.420	0.398	1.056×
tribonacci	0.420	0.401	1.048×
sum	0.399	0.382	1.046×
block	0.370	0.349	1.062×
loop_poly	0.369	0.349	1.058×
staged.brtable	0.383	0.357	1.074×
compare_wasp.large-branch	0.595	0.573	1.038×
compare_wasp.small-snapshot	0.406	0.383	1.059×
i64_ops	0.522	0.502	1.038×
call_indirect_test	0.421	0.399	1.055×
call_indirect_test.1	0.420	0.397	1.060×
data_sec	0.366	0.344	1.064×
long_loop	0.385	0.359	1.071×
Overall	16.117	15.380	1.048×

[Kraks]

Thanks for the update! The improvement looks marginal, but I think we should still remove Immer from the generated code dependency, which would allow us to try implement other data structures to replace it.