Add precompiled headers for framework and RELs

[caseif]

This PR adds a precompiled header (dolzel.pch) for use in RELs to match the original build process. This has a huge effect on weak function ordering and as a result many TUs are now matching as a result of these changes. This also seems to solve (at least partly) the Z2Calc::cNullVec, lit_1787, and lbl_*_data_10 .data entries and lit_1109-lit_1009 .bss entries seen across the codebase.

This PR also removes nosyminline on from the default REL flags (and from many individual framework objects) since it is no longer necessary in the majority of cases.

One issue with the changeset in its current state involves the handling of the sqrtf constants. The PCH needs to be built with .sdata and .sdata2 enabled in order for d_kankyo and d_resorce to match; however, this causes the constants to be placed in .sdata2 even in RELs such as in d_a_b_yo or d_a_L7op_demo_dr (where they should be in .rodata). This not only breaks matches, but also breaks linking entirely with this message:

#   Small data sections must have their own output sections specified in the 
#   linker command file.
#   No linker command file output section '.sdata2'.

The solution to this for now is to build the PCH twice: once with framework flags, and once with REL flags. Each object links one version or the other as appropriate. This may or may not be a fakematch and should be replaced if a better solution is found in the future.

This PR includes the changes from encounter/dtk-template#60 in order to support building the PCHs. objdiff isn't yet capable of understanding the weak data stripping going on here, so there are a LOT of false data regressions reported as a result.

See zeldaret/tww#860 for the corresponding Wind Waker PR for PCH changes.

[decomp-dev]

Report for GZ2E01 (8a30030 - ffd010d)

📈 Matched code: 79.56% (+0.02%, +2208 bytes)
📈 Linked code: 47.14% (+16.81%, +1930772 bytes)

✅ 5 new matches

Unit	Function	Bytes	Before	After
d_a_e_pm/d/actor/d_a_e_pm	cM3dGPla::~cM3dGPla()	+72	0.00%	100.00%
framework/d/d_camera	dCamera_c::lineCollisionCheckBush(cXyz*, cXyz*)	+18	89.64%	100.00%
d_a_e_rb/d/actor/d_a_e_rb	daE_RB_Create(fopAc_ac_c*)	+17	98.35%	100.00%
d_a_e_pm/d/actor/d_a_e_pm	daE_PM_c::SearchFarP()	+2	99.76%	100.00%
d_a_e_pm/d/actor/d_a_e_pm	daE_PM_c::At_Check()	+2	99.65%	100.00%

💔 3 broken matches

Unit	Function	Bytes	Before	After
d_a_obj_hata/d/actor/d_a_obj_hata	csXyz::~csXyz()	-60	100.00%	0.00%
d_a_obj_hata/d/actor/d_a_obj_hata	csXyz::csXyz()	-4	100.00%	0.00%
framework/d/d_map	dMap_c::getColor(int)	-1	100.00%	99.88%

[TakaRikka]

is the dummy necessary?

[caseif]

It's there to force it to be placed earlier in .rodata, not sure if there's a more elegant way to do this though.

[TakaRikka]

are these ones that were changed from matching to equivalent not compatible with this?

[caseif]

For d_map including the PCH completely messes up the function order - I think that this object may actually not have included it. For d_select_cursor I think the PCH is correct, but it messes up the vtable padding which was previously solved with a fakematch (which I think ultimately comes down to a vtable ordering issue). I'll revert the first change for now and add a comment that it might be fake - let me know if you want me to revert the second change as well.