Improve binding element type inference using CheckMode

[Babak K. Shandiz (babakks)]

Fixes #49989
Alternative to PR #50241.

Here, to prevent false circular-relationship on binding elements initialized with their siblings (like the code segment below), we've utilized CheckMode values other than Normal as an indicator of suppressing type-checking errors.

const [a, b = a] = [0];
// or
const [a, b = a + 1] = [0];

One gotcha in this PR was to revert the updated links (i.e., links.type) of a symbol. Because once a circular relationship error was encountered the type of the symbol is inferred to be any and then somewhere in the call stack, the symbol's associated links.type would be assigned to that returned any type. So, in cases other than CheckMode.Normal we'd replace back the links.type old value. (Sorry, that was not up-to-date)

The gotcha is that I had to suppress updating the symbol's links.type (to any type) for other than CheckMode.Normal cases, because that would block further efforts to find the correct type for the symbol. Now, I'm thinking maybe it's better to also check if the returned type is any and then do the suppression. I didn't find any better approach than this, so let me know if there's one:

function getTypeOfVariableOrParameterOrProperty(symbol: Symbol, checkMode?: CheckMode): Type {
    const links = getSymbolLinks(symbol);
    let result = links.type;
    if (!result) {
        result = getTypeOfVariableOrParameterOrPropertyWorker(symbol, checkMode);
        // For a contextually typed parameter it is possible that a type has already
        // been assigned (in assignTypeToParameterAndFixTypeParameters), and we want
        // to preserve this type.
        if (!links.type && !(checkMode && checkMode !== CheckMode.Normal)) {
            links.type = result;
        }
    }
    return result;
}

Further issues

Although this PR resolves common cases like the one above, but in cases like below false errors still occur. I didn't dig more to fix such issues because handling them would, in my opinion, clutter the type-checker without significant value in return due to the rarity of such usages. Please correct me if I'm wrong.

const [a, b = (() => 1 + a)()] = [0];

[Wesley Wigham (weswigham)]

I don't get why this'd be SkipContextSensitive, since that sets a lot of other stuff to skip work. Contextual would be the one that basically just means non-cacheable and no errors.

[Nathan Shively-Sanders (sandersn)]

Babak K. Shandiz (@babakks) Do you want to keep working on this? I'd like to close it otherwise to avoid stale PRs.

[Babak K. Shandiz (babakks)]

Nathan Shively-Sanders (@sandersn) I'll try to address Wesley Wigham (@weswigham)'s comment soon (before the weekend). Is that okay?

[Nathan Shively-Sanders (sandersn)]

Sure, no big rush, it's been months since I looked at this =)

[Babak K. Shandiz (babakks)]

Nathan Shively-Sanders (@sandersn) Since this PR was based on an old commit of the main branch, I re-applied the changes here (with a slight modification to restrict the side effects) on another PR, #56753. Please close this one and review the other when you deem fit.