fix: no keep alive for large payloads

[pauldambra]

we set keepalive to true for any request that is a POST

the fetch spec at https://fetch.spec.whatwg.org/#http-network-or-cache-fetch says

If contentLength is non-null and httpRequest’s keepalive is true, then:
If the sum of contentLength and inflightKeepaliveBytes is greater than 64 kibibytes, then return a network error.

read more at e.g. https://javascript.info/fetch-api#:~:text=But%20the%20keepalive%20option%20tells,for%20keepalive%20requests%20is%2064KB.

it is likely that a replay payload would be >64kb compressed

not all browsers obey this spec equally (of course) so it won't always fail, so it's not broken consistently enough to stand out.

but locally I can see network requests failing for this reason

fetch keepalive is intended for situations including those like analytics (👋 posthog) where you might want to allow a request to finish even though the page is navigating so we don't want to turn it off completely

lets estimate body size and set keepalive when its safe (which it should be for most analytics events)

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
posthog-js	✅ Ready (Inspect)	Visit Preview	Sep 15, 2024 2:39pm

[github-actions]

Size Change: +516 B (+0.04%)

Total Size: 1.21 MB

Filename	Size	Change
dist/array.full.js	346 kB	+129 B (+0.04%)
dist/array.js	162 kB	+129 B (+0.08%)
dist/main.js	163 kB	+129 B (+0.08%)
dist/module.js	162 kB	+129 B (+0.08%)

ℹ️ View Unchanged

Filename	Size
dist/exception-autocapture.js	10.4 kB
dist/recorder-v2.js	110 kB
dist/recorder.js	111 kB
dist/surveys-preview.js	59.8 kB
dist/surveys.js	66 kB
dist/tracing-headers.js	8.26 kB
dist/web-vitals.js	10.3 kB

_{compressed-size-action}

[daibhin]

Could abstract this but doesn't seem necessary given the comment explains a lot

Suggested change

	keepalive: options.method === 'POST' && (estimatedSize || 0) < 64 * 1024,
	keepalive: options.method === 'POST' && (estimatedSize || 0) < MAX_KEEPALIVE_KB,

[pauldambra]

yeah, i'm always torn by the tension between "clarify intent" and "have fewest parts"
if you make it a const it's arguably clearer but i need to look at multiple places to be sure i understand